Behaviour of Soil-Structure Interfaces Subjected To A Large Number of Cycles, Application To Pile

THÈSE
Pour obtenir le grade de

DOCTEUR DE L’UNIVERSITÉ DE GRENOBLE
Spécialite: Matériaux Mécanique, Génie Civil, Electrochimie
Arrêté ministériel: 7 août 2006
Présentée par
SURIYAVUT PRA-AI
Thèse dirigée par Pierre FORAY et

codirigée par Marc BOULON
préparée au sein du Laboratoire 3S-R

dans l’École Doctorale Ingénierie-Matériaux Mécanique Environnement
Energétique Procédés Production (I-MEP2)
Behaviour of soil-structure
interfaces subjected to a large
number of cycles. Application
to piles
Thèse soutenue publiquement le 28.02.2013,

devant le jury composé de:
M. Isam SHAHROUR
Professeur à Polytech Lille, Laboratoire Génie Civil et Géo-Environnement, (Rapporteur)
M. Eric VINCENS
Maı̂tre de conférence à l’Ecole Centrale de Lyon, Laboratoire LTDS, (Rapporteur)
M. Fabrice EMERIAULT
Professeur à l’INP Grenoble, Laboratoire 3S-R, (Examinateur)
M. Jean CANOU
Chercheur à l’ENPC, Laboratoire NAVIER-CERMES, (Examinateur)
M. Alain PUECH
Directeur Scientifique à FUGRO Geo Consulting, (Invité)
Directeur Technique ANR et Programme National SOLCYP
Acknowledgements
Grenoble, there have been so many things that I have learned and so many people I have
gotten to know. And this has made my life here memorable. Firstly, it has been my true
honor to work with Prof. Marc Boulon and Prof. Pierre Foray. They have always believed
in my potential. Without their visions, efforts and encouragements, this thesis could never
have reached completion. Especially Prof. Marc Boulon, I cannot find words to express
my gratitude for the helps in everything since I have been in Grenoble. My most heartfelt
thanks to Prof. Isam Shahrour, Prof. Eric Vincens, Prof. Fabrice Emeriault, Prof. Jean
Canou and Dr. Alain Puech, who serve as my thesis committee.
I would like to thank my officemates. They made my office a great work environment. My
special appreciation is extended to my colleagues (MEMS) and other people at laboratory
3S-R. Especially Mr.Alexandre Martin for his help and guidance regarding the correction
as well as the modification of the direct shear device. My gratitude also goes to Matias
Silva and Bram van den Eijden, my friends who have always helped me in solving several
computer problems. I would like to thank my Thai friends here at Grenoble, who have
made my life here very enjoyable.
My special appreciation goes to Ariyanan Pra-ai and Parnphan Pra-ai, my wife and daughter
who always stand beside me. Their love, strength, and wisdom have inspired me to be the
best I can be through many difficult times. Finally, I owe my deepest gratitude to my
parents, Inpan and Kanchana PRA-AI; their love is the greatest power in the world and
has driven me to succeed in every goal I have dreamed of.
Suriyavut PRA-AI January, 2013
i
Abstract
A series of monotonic and cyclic 2D direct shear tests on sand (Fontainebleau)-rough/smooth

material interfaces under constant normal load (CNL) and constant normal stiffness (CNS)
conditions are presented. The aim of these tests is to simulate the situation along the shaft
of piles subjected to a large number of cycles due to environmental or anthropic loadings.
These cycles (typically 104 ) are of small amplitude (10, 20 and 40 kPa in terms of shear
stress) as the service loads are not supposed to produce an early failure. These tests in-
clude the series of changing the cyclic amplitude in succession. The problem of loss of sand
between the box and the rough plate, typical phenomenon in this type of test, receives a
special attention. It is interesting to observe, according to the initial density, the position
of the ”center of cycles” in the stress plane (mean cyclic variables). Several factors such as
initial density (ID0 ), initial normal stress (σn cm0 ), level of initial mean cyclic stress ratio
(ηcm0 ), reduced cyclic amplitude (∆η) and imposed normal stiffness (in this thesis, k =
1000, 2000 and 5000 kPa/mm) that influence the intensity of mean cyclic normal ([u]cm )
and shear ([w]cm ) displacements are considered. Along CNL paths either dilation or con-
traction is exhibited, in agreement with the characteristic state developed by Luong. The
influence of the stress path (under constant normal stress or prescribed normal stiffness)
is also highlighted. It should be highlighted that CNS paths are ever contractive. The
model of monotonic interface behaviour under CNL and oedometer paths is fully analyt-
ical and based on the rate-type framework with the normal relative displacement or the
interface density as unique memory parameter. While the analytical formulations for iden-
tification are first proposed to describe the interface behaviour under cyclic CNL condition,
the variation of cyclic amplitude and CNS condition are modeled by applying the analytical
formulations for validation using finite analytical increments and introducing the notion of
equivalent number of cycles. For numerical simulations by the FEM (Plaxis) these tests are
interpreted and formulated according to a pseudo visco-plastic framework, the number of
cycles being a fictitious time. The direct shear tests and two centrifuge pile tests (pull-out
and compression) are also modeled. Recommendations are proposed for the calculation of
real usual piles under cyclic loading. This thesis is one part of the national French project
”ANR and National Program SOLCYP” (Research on behaviour of piles subjected to cyclic
loading).
KEYWORDS: Direct shear tests, granular soil-structure interface, cyclic loading, mean
cyclic path, large number of small cycles, characteristic state, pseudo visco-plasticity
ii
Résumé
On présente tout d’abord une série d’essais de cisaillement direct 2D monotones et cycliques
sur l’interface sable de Fontainebleau-plaque rugueuse et lisse, à contrainte normale con-
stante (CNL) et à rigidité normale imposée (CNS). Le but de ces essais est de simuler la
situation mécanique le long de pieux soumis à un grand nombre de cycles d’origine en-
vironnementale ou anthropique. Ces cycles (typiquement 10000) de faible amplitude (10
à 40 kPa en terme de contrainte de cisaillement) ne sont pas censés produire de rupture
prématurée. Ces tests incluent une série de cycles d’amplitudes (successives) variées. Le
problème de la perte de sable entre la boı̂te et la plaque est traité avec attention. Nous avons
interprété l’effet de la position du ”centre des cycles” dans le plan de contraintes (variables
cycliques moyennes) et de la densité initiale. Plusieurs facteurs tels que l’indice initial de
densité (ID0 ), la contrainte normale cyclique moyenne (σn cm0 ), le niveau initial moyen de
contrainte de cisaillement (ηcm0 ), l’amplitude cyclique reduite (∆η) et la rigidité normale
imposée (k qui dans cette thèse, va de 1000 à 5000 kPa/mm), influencent les déplacements
relatifs cycliques moyens normal ([u]cm ) et tangentiel [w]cm ) et sont pris en considération.
On observe soit de la dilatance, soit de la contractance en accord avec l’état caractéristique
développé par Luong. L’influence du chemin de contrainte (CNL ou CNS) est également
analysée. Un modèle phénomenologique et analytique de comportement d’interface sur
chemins cycliques CNL est proposé. C’est également le cas pour le comportement mono-
tone sur chemins oedométrique et CNL, la variable de mémoire unique étant la densité
d’interface (sous contrainte) ou le déplacement relatif normal. Cette formulation permet de
traiter, par incréments analytiques finis, les chemins comportant une variation d’amplitude
cyclique, et les chemins CNS, ce qui introduit la notion de nombre de cycles équivalent. On
notera que les chemins CNS sont toujours contractants. Ces essais sont utilisés pour aborder
la simulation par éléments finis, avec le logiciel Plaxis, selon une approche de pseudo-visco
plasticité, le nombre de cycles tenant lieu de temps fictif. L’essai de cisaillement monotone
à la boı̂te est modélisé en densités faible et forte, ainsi que deux essais de pieux modèles
centrifugés, l’un en traction, l’autre en compression. Des recommandations sont proposées
pour le calcul courant des pieux sous sollicitations cycliques. Cette thèse a été soutenue
par l’ANR SOLCYP et le programme national ” recherches sur le comportement des pieux
soumis à des sollicitations cycliques”.
Mots clés : essais de cisaillement direct, interface sol granulaire-structure, chargement cy-
clique, chemin cyclique moyen, grand nombre de petits cycles, état caractéristique, pseudo-
visco plasticité
iii
List of Symbols
• CNL Constant Normal Stress condition

• CNS Constant Normal Stiffness condition
• k [kPa/mm] Imposed normal stiffness
• τ [kPa] Shear stress
• σn [kPa] Normal stress
• τcm [kPa] Mean cyclic shear stress
• σn cm [kPa] Mean cyclic normal stress
• [u] [mm] Normal relative displacement
• [w] [mm] Shear relative displacement
• [u]cm [mm] Mean cyclic normal displacement
• [w]cm [mm] Mean cyclic shear displacement
• Rmax [mm] Maximum surface roughness
• Rn [-] Normalised surface roughness
• t [mm] Interface thickness
• [u]int [mm] Normal relative displacement of interface
• [u]sam [mm] Normal relative displacement of sample
• kn int [kPa/mm] Normal stiffness of interface
• kn sam [kPa/mm] Normal stiffness of sample
• γd0 σ n=0 [kN/m3 ] Specific weight of interface without an applied normal stress
• γd0 σn [kN/m3 ] Specific weight of interface under an applied normal stress
• γd [kN/m3 ] Current specific weight of interface during shear loading
• δpeak [◦ ] Peak friction angle of interface
• δcrit [◦ ] Critical friction angle of interface
iv
• δcar [◦ ] Characteristic friction angle of interface
• ηpeak [-] Peak stress ratio
• ηcrit [-] Critical stress ratio
• ηcar [-] Characteristic stress ratio
• ηcm [-] Mean cyclic stress ratio
• ηmax [-] Maximum stress ratio
• ∆τ [-] Cyclic amplitude in terms of shear stress
• ∆η [-] Reduced cyclic amplitude
• N [-] Number of cycles
v
Contents
Acknowledgements i
Abstract ii
Résumé iii
List of Symbols iv
General introduction 1
1 Bibliographic study of soil-structure interfaces subjected to cyclic loading 3

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Soil-structure interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Experimental investigation of interfaces . . . . . . . . . . . . . . . . . . . . 6
1.3.1 Interface shear devices . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.2 Interface roughness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.3 Interface thickness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Cyclic CNL interface shear tests . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 CNS interface shear tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5.1 Several shear paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5.2 Definition of Constant Normal Stiffness (CNS) . . . . . . . . . . . . 16
1.5.3 Cyclic CNS interface shear tests . . . . . . . . . . . . . . . . . . . . 19
1.6 Element tests on high-cycle accumulation . . . . . . . . . . . . . . . . . . . 25
1.6.1 Influence of the strain amplitude . . . . . . . . . . . . . . . . . . . . 28
1.6.2 Influence of initial density, ID0 . . . . . . . . . . . . . . . . . . . . . 29
1.6.3 Influence of of loading frequency . . . . . . . . . . . . . . . . . . . . 30
1.6.4 Influence of of the average stress . . . . . . . . . . . . . . . . . . . . 30
1.7 Constitutive models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.7.2 Interface constitutive models . . . . . . . . . . . . . . . . . . . . . . 35
1.7.3 Incremental models (Implicit in the sense of Wichtmann(2005)) . . . 38
1.7.4 Elastoplastic models (Implicit in the sense of Wichtmann(2005)) . . 41
1.7.5 Pseudo-creep models (Explicit in the sense of Wichtmann(2005)) . . 48
1.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
vi
2 Interface direct shear tests 55
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.2 Interface direct shear devices . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.3 Tested materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.3.1 Standard Fontainebleau sand . . . . . . . . . . . . . . . . . . . . . . 61
2.3.2 plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.4 Monotonic interface direct shear tests . . . . . . . . . . . . . . . . . . . . . 67
2.4.1 Interpretation of interface shear test . . . . . . . . . . . . . . . . . . 67
2.4.2 Test results of monotonic interface shear test . . . . . . . . . . . . . 71
2.5 Test parameters and extracted variables from the tests . . . . . . . . . . . . 91
2.6 Operating mode and experimental biases . . . . . . . . . . . . . . . . . . . . 93
2.6.1 Operating mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.6.2 Experimental bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.7 Identification tests (cyclic CNL tests on rough plate) . . . . . . . . . . . . . 98
2.7.1 Influence of density (ID0 ) . . . . . . . . . . . . . . . . . . . . . . . . 102
2.7.2 Influence of initial normal stress (σn ) . . . . . . . . . . . . . . . . . . 105
2.7.3 Influence of mean cyclic stress ratio ηcm . . . . . . . . . . . . . . . . 107
2.7.4 Influence of cyclic amplitude ∆τ . . . . . . . . . . . . . . . . . . . . 109
2.7.5 Post-cyclic interface behaviour . . . . . . . . . . . . . . . . . . . . . 112
2.8 Validation tests (cyclic CNS tests on rough plate) . . . . . . . . . . . . . . . 120
2.8.1 Influence of normal stiffness (k) . . . . . . . . . . . . . . . . . . . . . 120
2.8.2 Influence of ηcm0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
2.8.3 Influence of cyclic amplitude (∆τ ) . . . . . . . . . . . . . . . . . . . 132
2.9 Validation tests (cyclic CNL tests by changing cyclic amplitude) . . . . . . 137
2.10 Cyclic interface direct shear tests on smooth plate . . . . . . . . . . . . . . 141
2.10.1 Cyclic CNL tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
2.10.2 Cyclic CNS tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
2.11 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
3 Constitutive modeling of direct shear tests 157

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
3.2 Monotonic tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
3.2.1 Incremental (rate type) constitutive models . . . . . . . . . . . . . . 158
3.2.2 Analytical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
3.3 Cyclic tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
3.3.1 Modeling [u]cm path at constant normal stress (CNL) . . . . . . . . 176
3.3.2 Modeling [w]cm path at constant normal stress (CNL) . . . . . . . . 189
3.3.3 Checking of the proposed analytical models for identification . . . . 199
3.3.4 Validation of the identification (CNL) model to the variation of cyclic
amplitude and CNS condition . . . . . . . . . . . . . . . . . . . . . . 202
3.3.5 Checking of the proposed analytical models for validation . . . . . . 207
3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
4 Finite element modeling 214

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
vii
4.2 Direct shear box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.2.1 Monotonic CNL direct shear box . . . . . . . . . . . . . . . . . . . . 215
4.2.2 Monotonic CNS direct shear box . . . . . . . . . . . . . . . . . . . . 219
4.3 Centrifuge pile tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
4.3.1 Modeling of pile pull-out load test . . . . . . . . . . . . . . . . . . . 226
4.3.2 Modeling of pile pull-out load test (1st cycle) . . . . . . . . . . . . . 229
4.3.3 Modeling of pile pull-out load test (for further cycles) . . . . . . . . 232
4.3.4 Modeling of pile compression load test . . . . . . . . . . . . . . . . . 235
4.3.5 Modeling of pile compression load test (1st cycle) . . . . . . . . . . . 237
4.3.6 Modeling of pile compression load test (further cycles) . . . . . . . . 238
4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
5 Calculation of real piles subjected to cyclic loading, for civil engineering

projects 242
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
5.2 Behaviour of real piles subjected to axial cyclic loading . . . . . . . . . . . . 242
5.3 Determination of soil and interface parameters . . . . . . . . . . . . . . . . 243
5.4 Model prediction of piles under cyclic axial loading . . . . . . . . . . . . . . 244
5.4.1 Initial state (pile installation) . . . . . . . . . . . . . . . . . . . . . . 244
5.4.2 Characterizing the loading acting on pile head and along pile shaft . 244
5.4.3 Soil and soil-structure interface tests . . . . . . . . . . . . . . . . . . 244
5.4.4 FEM modeling the behaviour of a pile under axial cyclic loading . . 247
General conclusions and perspectives 248
Appendices 256
A Analytical model of monotonic interface tests 257

A.1 Detail of functions at before-peak phase . . . . . . . . . . . . . . . . . . . . 257
A.2 Detail of functions at post-peak phase . . . . . . . . . . . . . . . . . . . . . 258
B Results for Monotonic CNL and CNS tests 262

B.1 Monotonic CNL tests on loose sand (ID0 ≈ 30%) . . . . . . . . . . . . . . . 262
B.2 Monotonic CNL tests on dense sand (ID0 ≈ 90%) . . . . . . . . . . . . . . . 266
B.3 Monotonic CNS tests on loose sand (ID0 ≈ 30%) . . . . . . . . . . . . . . . 269
B.4 Monotonic CNS tests on dense sand (ID0 ≈ 90%) . . . . . . . . . . . . . . . 276
C Cyclic interface models 283

C.1 Constitutive interpolations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
C.2 Correlation between ηcm and ηcm2 . . . . . . . . . . . . . . . . . . . . . . . 285
C.3 Curve of pmax (∆η) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
D Cyclic CNL tests 288

D.1 Cyclic CNL tests on loose sand (ID0 ≈ 30%) with rough plate . . . . . . . . 288
D.2 Cyclic CNL tests on dense sand (ID0 ≈ 90%) with rough plate . . . . . . . 291
viii
E Cyclic CNS tests 293
E.1 Cyclic CNS tests on loose sand (ID0 ≈ 30%) with rough plate . . . . . . . . 293
E.2 Cyclic CNS tests on dense sand (ID0 ≈ 90%) with rough plate . . . . . . . . 296
F Résumé français étendu de la Thèse 301
ix
General introduction
Soil-structure interfaces are commonly encountered in many geotechnical engineering prob-

lems, e.g. shallow foundations, pile foundations, cutoff walls, earth reinforcement and buried
pipelines, tunnels etc. When civil engineering structures have to undergo cyclic loading con-
ditions, the bearing capacity of structures is often reductive. Especially, many offshore oil
rig works have to undergo cyclic loading conditions (wind, wave, machine operations, etc.)
for a long life. The recent developments of renewable installations at sea of energy sources
(offshore wind power, hydrokinetic power etc.) bring the professionals and the researchers
to be interested in the effect of very large number of cycles on the soil-structure interactions.
Understanding the interface behaviour subjected to cyclic loading is of significant impor-
tance. Indeed at the present time, there do not exist the methods of reliable calculation of
the structure foundations subjected to this stress type and most of the experts adopt the
proposed safety factors to take into account the degradations of bearing capacity due to the
cycles.
The proposed subject consists in working out a model of the behaviour taking into account
the effect of a large number of cycles on the elementary response of the soil and the soil-
structure interface and to introduce it into a computer code by finite elements in order to
be able to simulate the total behaviour of the works under this cyclic loading type.
The first part of this study will consist in developing a direct shear machine in order to
carry out cyclic soil-structure interface tests with controlled boundary conditions: constant
normal stress (CNL) and constant normal stiffness (CNS). Then the exhaustive test cam-
paigns to a large number of cycles, type of surface roughness, type of sand, variation of
densities, the initial normal stress condition, characteristic of shear path and characteristic
of mean cyclic shearing and the amplitude of shearing will be carried out.
From cyclic triaxial test results drawn from the literature and cyclic interface tests carried
out, the phenomenologic laws describing the evolution of the parameters of the soil with
the number of cycles will be worked out. From numerical point of view, the accumulation
of the irreversible quantities of cyclic origin (deformations, including volume in drained
condition, pore water pressure in undrained condition) will be treated like a cyclic pseudo-
creep, the number of cycles to be treated as a fictitious time. The established model, using
a pseudo visco-plasticity (Perzyna’s model) will also be tested on experiments conducted on
piles model, independently of this study, in laboratory and probably in the future for cyclic
loading tests of piles carried out on site. It is also interesting to simulate the phenomenon of
degradation of side friction along the piles during the application of large numbers of cycles.
1
This project is carried out within the framework of a national research project (framework
of IREX National Project) and ANR project (SOLCYP), ”Research on behaviour of piles
subjected to cyclic loading”.
The content of this thesis is organised with the objective mentioned above.
An analysis of the interface problems consisting in various experimental methods, emphasiz-
ing the contributions made in the past by different authors is presented in Chapter 1. The
element test results on high-cycle accumulation considered as a model for understanding
and modeling the interface test campaigns are also reviewed. The extension of the concepts
of constitutive laws based on the rate-type, elastoplastic and viscoplastic theories are then
summarized.
Chapter 2 contains the statement of the testing method and the device developed to im-
plement the interface shear tests. The description of tested materials is also discussed in
detail. Monotonic CNL and CNS tests serving as the preliminary decision for cyclic inter-
face shear tests are first outlined. Test parameters and extracted variables from the tests
including the operating mode and experimental biases are also described. Then, the overall
results of identification (CNL) and validation (CNL in changing cyclic amplitude and CNS)
tests are presented. The influences of initial density, initial normal stress, the level of mean
cyclic stress ratio and cyclic amplitude on the intensity of mean cyclic displacements are
also discussed.
Chapter 3 principally devotes to the constitutive modeling of interface direct shear. The
incremental (rate-type) constitutive model for monotonic soil-structure interface is first de-
scribed. Then the analytical models for identification (cyclic CNL condition) by using the
pseudo viscoplastic formulation (Perzyna’s model) in which the number of cycles is treated
instead of time are discussed in detail. The basic formulations used are analytical CNL
and oedometric paths depending on the cyclic variables (σn cm0 : mean cyclic normal stress,
ηcm0 : mean cyclic stress ratio, ∆η: cyclic amplitude, ID0 : density index). The integration
of the constitutive model along CNS path is realised by finite analytical increments (or
pre-integrated increments), with one single memory parameter: cyclic normal relative dis-
placement ([u]cm ) or interface specific weight under stress (γd σn ). Subsequently, the results
of the extension of analytical models for identification to validation of variation in cyclic
amplitude and CNS condition are presented.
Chapter 4 presents the preliminary implementation of Plaxis FE software for modeling
the interface behaviour. Monotonic interface direct shear tests (CNL and CNS) are first
modeled. Then the application of analytical constitutive models to centrifuge model test of
pile is presented. Monotonic pull-out/compression and cylic tests are typically described as
proposed methodology.
Finally, the proposed methodology for the calculation of real piles subjected to axial cyclic
loading is presented in Chapter 5.
2
Chapter 1
Bibliographic study of
soil-structure interfaces subjected
to cyclic loading
1.1 Introduction
In the field of soil mechanics applied to civil engineering, computing the bearing capacity,
as the stability of deep foundations and retaining structures, is largely based on the phe-
nomenon of skin friction between the soil and structure. This is the case of piles, anchors,
retaining walls, tunnels, etc as shown in Figure 1.1. The study of frictional characteristics
of the interface between granular materials and structure, developing through a thin layer
of granular material, therefore plays an important role for understanding the behaviour
of many geotechnical structures and also for providing more secure and cost savings in the
design of structures. This subject has widely been paid attention by many researchers. Gen-
erally, the studies of soil-structure interface have been developed in various categories (e.g.
fundamental and experimental studies, constitutive modeling, and application to boundary
value problems). The mechanisms and characteristics were commonly determined from ap-
propriate laboratory or field tests to describe the interface behaviour in a realistic manner.
This chapter begins by reviewing the state of the art on the experimental investigations in-
cluding the development of various testing devices on soil-structure interfaces. Afterwards,
necessary constitutive models formulated for interface behaviour will be described.
1.2 Soil-structure interface
Soil-structure interface is the result of a phenomenon induced by the discontinuity and

contrast of mechanical properties at the soil-structure contact. The interface behaviour
takes place in a very thin zone around the structure which develops a strain localization
caused by the transmission of tangential force of structure to the soil. A discontinuity of
3
mechanical properties has the important and irreversible changes in the configuration of less
resistant material during shear loading. For instance, the structure of granular medium is
completely altered to an extent depending on the level of existing stress. The interface zone,
consisting of one part soil and some particles torn from the structure, thus has mechanical
properties very different from the rest of the soil. Considering the soil-structure friction, its
characteristic is due to the presence of a static part accompanied by a kinematic part in the
expression of the contact law. Main observations and findings on the phenomenon based
on the existence of soil-structure interfaces show the shear resistance [τ ] deduced from the
classical formula as:
τ = σtanδ (1.1)
where δ is friction angle between soil and structure and σ is the normal stress exerted
on a structure anchored in the soil. The stresses and strains coupled with a variation of
deformations in the immediate vicinity of the structure are only meaningful in a zone of
limited thickness around the inclusion. The methods for predicting the bearing capacity of
piles and the laws of mobilization of skin friction along the shafts have also been proposed.
(a) (b)
Figure 1.1: Application of soil-structure friction; (a) earth retaining structures with geosyn-
thetic reinforcement; (b) pile foundations
In most cases involving geotechnical offshore, the bearing capacity of piles are commonly
designed by following the guidelines of the American Petroleum Institute API [1993]. The
Coulomb failure criterion used to calculate the shaft friction of pile is defined as;
τs = Kσn′ tanδ (1.2)
where K is an earth pressure coefficient relating the effective normal stress acting around
the piles at failure to the in situ effective over burden stress (σv′ ), tan δ is the friction
coefficient between pile and soil. The value of K(≈ 0.8 − 1) is assumed constant along the
4
pile shaft. API method would give over or underestimation because the density of the soil
surrounding the pile is not taken into account. However, this method still provides the
reasonable predictions for the dimensions of piles and soil types of the used database.
Schlosser & Guilloux [1981] defined a friction coefficient, so called apparent, as a function
of shear stress during the pull-out test on the strip related to the initial normal stress by
the following equation;
µ∗ = τ /σ0 (1.3)
with µ∗ ≥ µ = τ /(σ0 + δσ) ; µ is the real soil-structure friction coefficient and ∆σ is the
variation of normal stress during the pull-out test.
P*
tan I
tanI
P
Vn (kPa)
(a) (b)
Figure 1.2: Dilatancy effects in pull-out tests on strips; (a) influence of soil dilatancy on
the friction coefficient; (b) local increase of normal stress as a result of dilatancy, Schlosser
& Guilloux [1981]
From the experimental observations, it was commonly found that the values of the apparent
friction coefficient (µ∗ ) were much higher than the real friction coefficient. The high value
of the apparent soil-structure friction angle and the increase in normal stress, were due
to dilatancy. The expansion or contraction of the soil caused by the shear loading is a
phenomenon which requires the comprehension of the evolution of the interface mechanical
properties. Figure 1.2 shows that a pull-out test of an inclusion embedded in the soil
produces the shear stress whose values are significant only in a limited zone around the
inclusion. In this area the soil tends to increase in volume due to expansion, but is partly
prevented by the rest of the ground. This causes a significant increase of normal stress
and consequently the friction coefficient. The roles played by the factors influencing the
behavior of interface such as roughness and density have an effect on dilatancy.
5
1.3 Experimental investigation of interfaces
So far, the development of accurate understanding of the mechanical response of interfaces

has alternatively been investigated from appropriate laboratory or field tests to describe the
interface behaviour in a realistic manner. A comparison was made of their advantages and
disadvantages with respect to the others. However, a number of refinements and modifica-
tions of several devices have continuously been made in order to enhance the comprehension
of interface behaviour.
1.3.1 Interface shear devices
One of the most extensive studies on soil-structure interface testing has been performed by
using the conventional direct shear devices whether at constant normal load (CNL) or at
constant normal stiffness (CNS) condition. Figure 1.3 shows typical direct shear test which
is widely used to investigate the interface behaviour (Desai et al. [1985]; Boulon [1989];
Hoteit [1990]; Al-Douri & Poulos [1992]; Tabucanon et al. [1995]; Mortara [2001]; Shahrour
& Rezaie [1997]) as a result of much less technical difficulty and commonly available devices.
Nevertheless, the problem arising during the test is that the stress and strain states within
the sample are not homogeneous due to the effect of side wall friction. The stress non-
homogeneity arises as a consequence of shear loading which is transferred through the wall
of the shear box. On the condition that the interface test considered as a surface test (due
to the very intense localization) instead of a volume test, therefore the homogeneity is only
taken into account along the idealized displacement discontinuity surface. The shear band
occurring by using direct shear test still remains at the place where it took place.
Boulon [1989] also performed a series of direct shear tests on sand-structure interfaces
and proposed the idealization of soil structure interfaces. From this investigation the soil-
structure interface behaviour was practically considered as a surface (or contact with fric-
tion) behavior which occurred between the two boundaries of a shear band inside the soil
along the structure in bidimensional situation. A shear band or soil-structure interface
cannot explicitly be separated from the surrounding soil, for this reason, the direct shear
test has to be interpreted by considering two parts in the sample. Figure 1.4 also illustrates
the direct shear test interpretation: the ”active” lower part being in contact with the rough
material, so called interface whose thickness is about 10 times the mean diameter D50 of the
sand grains, and the ”passive” upper part as thin as possible, acting as oedometer sample.
The interface variables taken into account are then vector variables: the stress vector acting
on the interface (τ ′ , σn′ ): ′
τ τ
σ= ≈ (1.4)
σn′ σn
And the relative displacement vector between the two boundaries of the interface, where
[u′ ] and [w] are normal and tangential displacements of interface respectively:

[w] [w]
[u] = ′ ≈ (1.5)
[u ] [u]
6
Figure 1.3: Direct interface shear apparatus, Tabucanon et al. [1995]
All the components have to be considered as mean values within the generated contact
surface. In addition, the roles played by the factors influencing the behaviour of interface
from various observations are roughness, normal stress level, constraint conditions, grain
strength and density.
Figure 1.5 shows simple shear apparatus which can also be performed with much less diffi-
culty in which the tangential displacement can be measured separately in terms of sliding
displacement of the interface and displacement due to shear deformation of sand (Uesugi
& Kishida [1986]; Kishida & Uesugi [1987]; Fakharian & Evgin [1997]). This is able to
overcome the shortcoming of direct shear box. The simple interface consists of a stack of
rectangular plates as the sample container placed on the plate. There are still the effect of
side wall friction as well as the friction between the sample container and the plate. More-
over, this should also be taken into account the shortcomings as material loss occurring
between the shear box and the plate during testing. The leakage of sample would lead to
an accuracy of normal displacement. Various solutions are then used, for instance, a pair
of thin steel plates inserted between the top and the bottom halves of direct shear box
(Al-Douri & Poulos [1992]) and thin layer of foam and Teflon sheet pasted between the
plate and the shear box (Fakharian & Evgin [1997]), were subsequently used in order to
prevent the loss of material during testing.
7
(a) (b)
Figure 1.4: Basic features of soil structure interface behavior: (a) Interface idealization; (b)
Direct shear test interpretation, Boulon [1989]
Figure 1.5: Simple shear type friction apparatus and simplified view of laterally constrained
sample under simple shear conditions, Fakharian & Evgin [1997]
The ring shear would be the most efficient device to observe the interface behaviour (Yoshimi
& Kishida [1981], Kelly [2001]). Because the sample is ring-shaped (i.e. there is unlimited
distance for shearing), it is free from the progressive failure and the contact surface still
remains constant whether the sample is sheared at a large circumferential displacement.
The stress states are homogeneous through shear test. However, some disadvantages still
exist, since the feature of this device is ring-shaped, it seems inconvenient to prepare the
sample due to more technical difficulty. A variety of ring shear tests are presented in Figure
1.6. The sample is placed in the container below the loading ring and is sheared against the
surface of loading ring. The normal load is applied through the top platen and the shear
stress is determined from the applied torque.
8
1.3.2 Interface roughness
From experimental investigations, it has obviously been found that surface roughness of
structures is a major factor affecting soil-structure friction coefficient. It was primarily
quantified as a maximum height Rmax (Yoshimi & Kishida [1981]). The roughness Rmax (L
= 2.50 mm) was defined as the relative height between the highest peak and the lowest along
a surface profile over 2.5 mm length. The diameter of sand particle was not considered in
this definition (Figure 1.7). However, in order to correlate the surface roughness with the
coefficient of interface friction, the diameter of sand particle should then be considered in
evaluating the surface roughness. The diameter of sand particle can be incorporated into
Rmax by modifying the gauge length L to the 50% mean diameter of sand particle (D50 ).
The modified roughness was represented in terms of Rmax (L = 0.20 mm) (Kishida & Uesugi
[1987]), as shown in Figure 1.8. In evaluating the surface roughness of steel, a normalized
roughness Rn was then defined as
Rmax (L = D50 )
Rn = (1.6)
D50
where Rmax (L = D50 ) is the Rmax value of a steel surface with gauge length L = D50 .
Figure 1.6: Ring shear device (left) Yoshimi & Kishida [1981], (right) Kelly [2001]
9
L = 2.5 mm.
Figure 1.7: Evaluation of surface roughness and modification of surface roughness evalua-
tion, Kishida & Uesugi [1987] and Uesugi & Kishida [1986]
D50
Rmax
Rough plate
L = D50
D50
Rmax
Smooth plate L = D50
Figure 1.8: Simple interpretation of interface roughness; rough and smooth surface, Uesugi
& Kishida [1986] and Mortara et al. [2007]
The normalized roughness is used to distinguish the type of the roughness in terms of
smooth interface or rough interface. The value of critical roughness (Rcrit ) is therefore used
to classify these terms of the interface. Available investigations (Uesugi & Kishida [1986],
Uesugi et al. [1989], Hu & Pu [2004]) indicate that the critical roughness (Rcrit ) is defined
in the range of 0.1-0.13, i.e. Rn < Rcrit (smooth interface) and Rn > Rcrit (rough interface).
Figure 1.9 also shows typical influence of the normalized roughness. As can be seen in this
figure, the peak shear stress is a function of normalized roughness. The dilative behaviour
can be observed on the test with high value of normalized roughness.
10
200
sheear stress at contact-face /kPa

150
100
50
0
0 2 4 6 8 10 12
relative displacement at contact-face /mm
1.2 R = 0.01
R = 0.05
1.0
normal displacement /mm
R = 0.1
0.8 R = 0.2
R = 0.5
0.6
0.4
0.2
0.0
0 2 4 6 8 10 12
-0.2
relative displacement at contact-face /mm
Figure 1.9: Influence of roughness surface (R ; see definition 1.6 hereabove) on interface
CNL sand (silica)-structure tests (Dr = 90%, σn = 200 kPa), Hu & Pu [2004]
1.3.3 Interface thickness
The evaluation of the thickness of the soil-structure interface has been the subject of sev-
eral studies carried out using different methods. It is worth noting that to define the exact
interface thickness is quite delicate and requires a high quality measurement. For instance,
Hoteit [1990] evaluated the interface thickness by using the stereophotogrammetry while
Dejong et al. [2003] investigated the shear deformation and volume change taking place
within the shear band by using particle image velocimetry (PIV). However, the parame-
ters discussed and identified by different authors have qualitatively the same influence. In
general, the grain size, as well as the roughness surface, influences the interface thickness.
The grain size influences the interface thickness as an increasing function of average grain
diameter. From several observations, the range of 7-14D50 is generally taken into account
for estimating interface thickness.
The thickness of the shear band tends to decrease when the initial density increases. Many
experimental results show that the relative displacements as well as the rotations of grains
in loose sand-structure interface are higher than that in the case of dense sand (Figure
1.10). This is explained by the fact that the grains of loose sand have a greater freedom of
11
movement than that of dense sand. From the investigation carried out by Hammad [1991],
the interface thickness for loose sand was in the range of 6-7D50 while for dense sand the
range of 4-5D50 was observed with rough interface.
Figure 1.10: Thickness of shear bands relative to the grain size (D50 ) as a function of
confining stress (σn ), Hammad [1991]
1.4 Cyclic CNL interface shear tests
The development of cyclic testing has widely been conducted on sand-structure interfaces,
whether by using modified direct shear tests or simple shear tests on various sands. Several
investigations have been carried out by varying amplitude of displacement, amplitude of
shear stress and normal stress, number of loading cycles, initial relative density of sand and
surface roughness (Desai et al. [1985]; Al-Douri & Poulos [1992]; Fakharian & Evgin [1997];
Mortara [2001]; Shahrour & Rezaie [1997]; Oumarou & Evgin [2005]).
Figure 1.11 and Figure 1.12 show monotonic and cyclic interface shear tests, on loose
(Dr = 35%) and dense (Dr = 85%) Toyoura sand, respectively, performed by Mortara
[2001]. The applied cycles are symmetric in terms of shear displacement. For smooth inter-
face, it could be seen that the variation on peak shear stress was not significant whether the
interface was subjected to cyclic loading on loose sample or dense sample. Considering the
evolution of normal displacement, the contracting interface behaviour (u>0) was observed
on both samples. On the other hand, rough interfaces exhibited the increase of maximum
shear resistance mobilized in the post cyclic phase providing this is tested on dense sam-
ple. Moreover, a contracting behaviour followed by dilation at each stress reversal on dense
sample was significantly observed. The contractive accumulation rate decreased with an
increase of the number of cycles. In addition, the post cyclic behavior exhibited an obvious
dilation because cyclic loading at constant normal stress leads to the progressive densi-
fication of the interface and the dilatancy angles consequently increase (Mortara [2001];
Shahrour & Rezaie [1997]).
12
Figure 1.11: Comparison between monotonic and cyclic CNL interface shear tests on smooth
and rough surface, Toyoura (silica) sand (Dr = 35%, σn = 150 kPa) u > 0 contraction,
Mortara [2001]
The amplitude of cyclic loading also influences the behaviour of interfaces. Figure 1.13
shows the intensity of normal displacements of interface subjected to cyclic loading with
different amplitudes of cycles (A) on loose Toyoura sand (Dr = 35%). With increasing
amplitude of cycles (in the range of A = 0.4-0.8 mm), the normal displacement tended to
be more contractive (Mortara [2001]).
1.5 CNS interface shear tests
1.5.1 Several shear paths
The study of interface behaviour has been specified in the conventional interface experiments
in which the applied normal stress is controlled to be constant (CNL) during the process
of shearing. Especially when interfaces subjected to cyclic loading, numerous experimen-
tal investigations have been reported that the interface responses turn into a progressive
densification with increasing number of cycles and stiffening rate decreases as a function of
number the cycles. This leads to the mobilized shear stress (Desai et al. [1985]; Al-Douri &
Poulos [1992]; Shahrour & Rezaie [1997]; Oumarou & Evgin [2005]).
13
Figure 1.12: Comparison between monotonic and cyclic CNL interface shear tests on smooth
Mortara [2001]
Nevertheless, from pull-out test of a model-pile embedded in sand performed by Puech et

al. [1979] the results obtained indicate that the variations of shaft friction coefficients (K)
during the loading correspond to the tendency of soil close to the pile whether to dilate or
to contract during the localized shear. For instance, the dilatancy is partially impeded by
the surrounding soil mass and then leads to an increase in lateral stress. Boulon & Foray
[1986] performed cyclic pull-out test of model-pile in sand. From this investigation, the
skin friction decreased as a function of number of cycles as shown in Figure 1.14. This
can be attributed that cyclic loading induces a contraction of the sand adjacent to pile. A
decrease in volume of sand leads to a progressive decrease in lateral stress, and consequently
a decrease of shear resistance.
Boulon & Foray [1986] also proposed the boundary conditions of direct shear tests in which
the stiffness of surrounding soil was represented by a pressuremetric modulus in order
to simulate the elementary mechanism of mobilization of lateral friction at the pile-soil
interface. This boundary condition, so called constant normal stiffness (CNS), has been
situated between the two extreme conditions, first the constant normal load (CNL) as
mentioned before. The other assuming no-volume change has not been realistic because of
considering the mass of the surrounding soil beyond the sheared zone as radially perfectly
rigid. Figure 1.15 shows how the interface behaved under two extreme conditions at high
14
and low initial normal stress respectively. The obvious difference between the two tests can
be observed, especially the mobilized normal stress on constant volume test.
Figure 1.13: Influence of different amplitude on cyclic CNL interface shear tests on smooth
Mortara [2001]
3
W (kPa)
0
0 50 100 150 200 250
Number of cycles (N)
Figure 1.14: Shear stress degradation with increasing number of cycles on cyclic pull-out
test of model-pile embedded in Hostun (silica) sand; mean unit skin friction, Boulon &
Foray [1986]
15
(a) (b)
Figure 1.15: Extreme interface direct shear tests on Hostun (silica) sand; (a) at low initial
normal stress σn = 12 kPa on dense sample; (b) at high initial normal stress σn =1061 kPa
on loose sample, Boulon [1989] and Boulon & Foray [1986]
1.5.2 Definition of Constant Normal Stiffness (CNS)
A laboratory test of pile-soil interface can simply be interpreted as an interface under

constant normal stiffness (CNS) condition. Considering a pile with radius R0 , embedded in
soil with a pressuremetric modulus Ep , and the thickness of the interface layer (e) mobilized
during the large localized shear (e << R0 ). The normal stiffness imposed to the interface
(k) according to Boulon & Foray [1986] results from the definition of the pressuremetric
modulus Ep (Figure 1.16, ∆V /V0 = relative volume change):
∆σ ∼ ∆σ ∆σ
Ep = = (R0 + e) ∼
= (R0 ) (1.7)
−2∆V /V0 −2∆[u] −2∆[u]
Consequently;
∆σ 2Ep
= =k (1.8)
−∆[u] R0
where ∆σ represents the variation of normal stress acting on the interface, ∆[u] the variation
of normal displacement.
Figure 1.16 shows the analogy between the localized shear along a pile and a direct shear
test with an imposed normal stiffness. In this case the normal stress acting on the interface
was applied by a spring with and appropriate stiffness. The values of the stiffness can be
distinguished in three typical boundary conditions.
• case I: k = 0, ∆σ = 0, ∆[u] 6= 0(constant normal stress)
16
• case II: k = ∞, ∆σ 6= 0, ∆[u] = 0(constant volume)
• case III: k = constant, ∆σ 6= 0, ∆[u] 6= 0(constant normal stiffness)
For instance, in case I, the normal stress applied on the interface is controlled to be constant
throughout shearing. For case II, the volume of the sample remains constant during shear-
ing. Case III in which the normal stress and the normal displacement vary proportionally
to ∆σn /∆u= k, as shown in Figure 1.17.
Boulon & Foray [1986] also illustrated a comparison of direct shear tests with imposed
stiffness (k) existing intermediately between the extreme conditions as shown in Figure
1.18. Furthermore, grading characteristics of sand and crushability of sand grains could
also been found to be the major factors on interfaces. Figure 1.19 shows a comparison
between interface direct shear tests under CNS condition performed on siliceous and on
calcareous sands. For a given stiffness, the mobilization of normal stress associated with
shear stress observed on calcareous sand is lower than that on siliceous sand, as a result of
the high crushability of sand grains on calcareous.
Figure 1.16: Analogy between the localized shear along pile and a direct shear test with an
imposed normal stiffness, Boulon & Foray [1986]
Vn = constant
Fixed Boundary
soil soil soil
structure structure structure

Fixed Boundary Fixed Boundary Fixed Boundary
k=0 k= k = constant
constant normal stress constant volume constant normal stiffness
Figure 1.17: Boundary condition in direction normal to interface, Fakharian & Evgin [1997]
17
1.00 0.60
constant normal stress constant normal stress
k = 233 kPa/mm k = 40
233kPa/mm
kPa/mm
0.50 k = 233
833 kPa/mm
0.80 k = 833 kPa/mm
isochoric
isochoric
0.40
0.60
W (MPa)
W (MPa)
0.30
0.40
0.20
0.20
0.10
0.00 0.00
0 2 4 6 8 10 12 0 2 4 6 8 10 12
[w] (mm) [w] (mm)
(a) (b)
Figure 1.18: Hostun (silica)sand-structure interface direct shear tests with three different
boundary conditions; (a) dense samples; (b) loose samples, with σn0 =354 kPa, Boulon &
Foray [1986]
Using CNS direct shear apparatus to investigate the behaviour of soil-structure interface,
many researchers have investigated the effect of different constant normal stiffness on the
response in both static and cyclic conditions and the variation of interface roughness to
cover the whole range of soil-structure interfaces.
Johnston et al. [1987] used CNS direct shear testing techniques to investigate the factors
which could control the behavior of rough concrete-rock interfaces and to develop more
rational methods for the prediction of side shear resistance in rock-socketed piles (Figure
1.20). This observation can explain that the dilation of the socket causes an increase in
the stress acting normal to the pile shaft and the relative movement between the pile
and the rock mass is controlled by CNS conditions as opposed to the more conventionally
encountered conditions of constant normal load (CNL) referring to the at rest K0 state.
Several monotonic CNS direct shear tests performed, for instance, by Hoteit [1990]; Tabu-
canon et al. [1995] and Mortara [2001] obviously showed the influence of the stiffness on
interface behaviour. Figure 1.21 shows that the larger CNS value on rough/dense sample
could cause a mobilization of the stresses acting on interface. It can be said that during
shearing phase, the interface exhibited a dilation and then led to an increase in normal
stress associated with shear stress. The normal displacements reduced for increasing values
of the normal stiffness. On the other hand loose samples showed a significant degradation
of the stresses as a result of the contractive behavior of interfaces. For smooth interface, the
peak shear stresses were very low due to the normal stress decreased rapidly. The normal
stiffness had little influence on the response as shown in Figure 1.22. From this observation,
the normal stress could vanish for small shear displacement (15-20 mm.), depending on
the conditions. This is due to the fact that smooth interface always exhibits contractive
behavior and sliding then occurs along the contact surface between sand-structure (Uesugi
& Kishida [1986]).
18
0.80 0.50
siliceous ; k = 833 kPa/mm siliceous ; k = 233 kPa/mm
siliceous ; k = 233 kPa/mm siliceous ; k = 40 kPa/mm
0.40 calcareous ; k = 233 kPa/mm
calcareous ; k = 833 kPa
0.60 calcareous ; k = 40 kPa/mm
calcareous ; k = 233 kPa/mm
W (MPa)
0.30
W (MPa)
0.40
0.20
0.20
0.10
0.00 0.00
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.10 0.20 0.30 0.40 0.50 0.60
Vn (MPa) Vn (MPa)
(a) (b)
Figure 1.19: Sand-structure interface direct shear tests with imposed stiffness on siliceous
and calcareous sand; (a) dense samples; (b) loose samples, with σn0 =354 kPa, Boulon &
Foray [1986]
(a) (b)
Figure 1.20: Behaviour of pile socketed in rock (a) displacement behavior; (b) principle of
the CNS test technique, Johnston et al. [1987]
1.5.3 Cyclic CNS interface shear tests
The effects of normal stiffness (k) on the cyclic interface behaviour performed by either
direct shear or simple shear tests are more significant than that on monotonic interface be-
havior. The main characteristic of cyclic interfaces under CNS condition is the degradation
of normal stress accompanied with the degradation of shear stress as a result of the progres-
sive densification. Figure 1.23 shows the typical results of cyclic CNS tests on dense/rough
interface performed on direct shear test by Airey et al. [1992].
19
Figure 1.21: Interface test results for different values of the normal stiffness. Dense Toyoura
(silica) sand,(Dr = 85%, σn = 200 kPa, rough surface Rmax = 60µm), u > 0 contraction
Mortara [2001]
160 80
k =1000 kPa/mm
k =1000 kPa/mm
K = 5000 kPa/mm
K = 5000 kPa/mm
120 60
Vn (kPa)
W (kPa)
80 40
40 20
0 0
0 5 10 15 20 25 0 5 10 15 20 25
[w] (mm) [w] (mm)
Figure 1.22: Interface test results for different values of the normal stiffness on loose Cal-
careous sand-smooth surface, σn = 125 kPa, Hoteit [1990]
20
Figure 1.23: Typical cyclic CNS shear test with k = 1600 kPa/mm on dense calcareous
sand/rough surface, σn = 250 kPa, Airey et al. [1992]
Fakharian & Evgin [1997] also reported the effect of magnitude of normal stiffness on
dense/rough interface by means of simple shear test. From this experimental investiga-
tion the normal and shear stresses decreased rapidly as a function of N with increasing the
normal stiffness (k) as shown in Figure 1.24. The amplitude of cycles also influenced the
stress gradation. If the amplitude of tangential displacement was large enough (0.75mm,
in this observation) to be able to reach the peak stress ratio with certain cycles, then a
degradation of stresses led to the failure of interface as shown in Figure1.25(a). The stress
path illustrated in Figure 1.26 also shows a gradual decrease in normal stress associated
to shear stress with cycles in case of cyclic amplitude = 0.75 mm. The stress path cycles
evolved inside the peak envelope before the failure. After touching the failure envelope
(cycle 11) the stress path subsequently bounced back inward and stabilized on the residual
envelope. For small amplitude of tangential displacement (0.25 mm), Figure 1.25(b), the
reduction of normal stress associated with small reduction in maximum shear stress could
hardly lead to failure. The failure of interface did not occur if the sliding displacement did
not accumulate with cycles.
21
Figure 1.24: Variation of normal and shear stress during cyclic CNS shear test with different
values of k (prescribed normal stiffness) on dense silica sand/rough surface, σn = 300 kPa,
Fakharian & Evgin [1997]
(a) (b)
Figure 1.25: Variation of normal and shear stress during cyclic CNS shear test with different
amplitudes of cycles on dense silica sand/rough surface, σn = 300 kPa; (a) 0.75 mm; (b)
0.25 mm , Fakharian & Evgin [1997]
22
Figure 1.26: Stress path touching the failure (N=11) during cyclic CNS shear test with
amplitudes of cycles 0.75 mm on dense silica sand/rough surface, σn = 300 kPa, Fakharian
& Evgin [1997]
Figure 1.27: Shear and normal stress degradation on cyclic CNS interface tests on Toyoura
(silica) sand with rough surface, σn = 150 kPa, Mortara et al. [2002]
Mortara [2001] and Mortara et al. [2002] also reported an important influence of normal
stiffness on cyclic interface especially on loose sample. Figure1.27 shows the significant
degradation of normal and shear stresses with an increase of normal stiffness on loose
23
sample. In this case, on rough interface with k = 1000 kPa/m three cycles were sufficient
for the normal stress dropped rapidly to a minimum value. On the other hand dense samples
exhibiting low tendency to contract required a higher number of cycles to reach the critical
condition.
The effect of normal stiffness (k) on different roughness has also been reported by Mortara
et al. [2007] as shown in Figure 1.28. The alternating phases of compression followed by
dilation were significant on rough interface. In this case, the overall behaviour still exhib-
ited contraction and therefore normal and shear stresses decreased. Furthermore, the post
cyclic phase showed a recovery of normal and shear stresses as a result of dilative behaviour.
On the other hand smooth roughness showed the progressive contraction without dilative
phase. For this reason there was no recovery of stresses in post cyclic phase.
shear stress, W (kPa)
rough surface smooth surface
shear displacement, [w] (mm) shear displacement, [w] (mm)

normal stress, Vn (kPa)
normal stress, Vn (kPa)
shear displacement, [w] (mm) shear displacement, [w] (mm)

normal stress, Vn (kPa) normal stress, Vn (kPa)
Figure 1.28: Comparison between cyclic shear displacement controlled CNS tests with dif-
ferent values of roughness on Toyoura (silica) sand, σn = 150 kPa, Mortara et al. [2007]
24
1.6 Element tests on high-cycle accumulation
The element test results performed on triaxial tests will be presented in the following sec-
tions. These are not related to interfaces but to volume behaviour. However, these very
interesting results should be considered as a model for understanding and modeling a part
of cyclic interface tests. Boulon & Nova [1990] illustrated the similarity in the interface be-
haviour exhibited with respect to sand behaviour on triaxial path. It is obvious to describe
that the terms of deviatoric stress (q), deviatoric strain (ε1 ), confining pressure (p’) and
volume strain (εv ) from triaxial paths are respectively substituted by the terms of τ , [w],
σn and [u] from interface direct shear tests as shown in Figure 1.29.
Considerable deformations in soils due to the accumulation of irreversible strain under the
effect of cyclic loading can lead to many problems. Even though the amplitude is very
small (e.g. εampl < 10−3 , Wichtmann [2005]) but a great number of cycles (e.g. N > 103 )
would lead to significantly harmful phenomena. In drained condition cyclic loading leads
to the accumulation of irreversible deformation either in contraction or in dilation and the
modification of soil modulus due to variation of density. Therefore, these issues are of
practical importance.
In this study the term ”accumulation” is used in such a way that it means either an increase
or decrease of state variables. A cyclic loading is able to lead to either residual strains or
change of stress, depending on the boundary conditions applied. Figure 1.30 shows typical
accumulation phenomenon under cyclic loading. If stress cycles are applied, this can lead
to an occurrence of cyclic pseudo-creeps (Figure1.30a) and if strain cycles applied then a
phenomenon of stress-relaxation can be observed (Figure 1.30b). Both conditions, pseudo-
relaxation and pseudo-creep, may occur simultaneously in the case of mixed-controlled tests
(Figure 1.30c).
Most of ideal approaches should be capable of simulating all the features of soil behaviour.
There are two main categories for calculating the accumulation of stress and strain caused
by cyclic loading: the ”implicit” method and the ”explicit” method.
The implicit or incremental method requires several (100) increments per cycle which may
lead to an accumulation of unavoidable numerical error and inaccurate results for a high
number of cycles, in general this method is appropriate for N < 50.
Thus it is obvious that the importance of explicit models is proposed, especially for a high
number of cycles, in such a way that only one or the first few cycles are calculated using
an incremental computation, and the rest of cycles will be treated as a cyclic pseudo-creep.
The first cycle whose shape is irregular may differ from the following ones. Therefore in this
method the accumulation is determined from the second cycle in which the number of cycles
(N) is taken into account instead of time (t) as shown in Figure 1.31(a). Several studies
have been conducted in the process of establishing patterns of cyclic behavior of materials
and especially to estimate the accumulated strain (pseudo-creep), which also causes an
accumulation of stress.
25
(a) (b)
Figure 1.29: Similarity in sand behavior on triaxial paths exhibited with respect to the
interface behavior; (a) triaxial tests; (b) interface direct shear tests, Boulon & Nova [1990]
Figure 1.30: Accumulation of stress or strain illustrated for the two-dimensional case; (a)
strain accumulation due to stress cycles applied; (b) stress accumulation due to strain cycles
applied; (c) mixed control tests, Wichtmann [2005]
26
(a) (b)
Figure 1.31: (a) Basic idea of explicit calculation strategies of the cumulative deformation;
(b) State of stress in cyclic triaxial test , Wichtmann [2005] and Wichtmann et al. [2005]
Wichtmann [2005], Wichtmann et al. [2005] and Wichtmann et al. [2010] performed several
series of drained tests with triaxial compression and uniaxial stress cycles with large number
of cycles (103 − 106 ) and relative small amplitudes. Figure 1.31 illustrates the state of stress
in cyclic triaxial test in which;
• The effective Cauchy stress σn (positive for compression) are used according to Roscoe
invariants p (mean pressure) and q (deviatoric stress) as:
p = (σ1 + 2σ3 )/3 (1.9)
q = (σ1 − σ3 )/ (1.10)
• The strain in axial direction is denoted with ε1 and the one in radial direction ε2 =
ε3 . The strain invariants can therefore be expressed as:
εv = ε1 + 2ε3 (1.11)
εq = 2/3(ε1 − 2ε3 ) (1.12)
• The critical friction angle is denoted by ϕc and the other one at peak as ϕp .
27
• The critical state lines in compression and extension tests are denoted by Mc and Me
as:
6 sin(ϕc ) 6 sin(ϕe )
Mc = and Me = (1.13)
3 − sin(ϕc ) 3 − sin(ϕe )
• The average mean stress pav was kept constant and the axial stress component σ1 was
cyclically varied with amplitude σ1ampl . The normalized size of the stress amplitude
can be described as the ratio of:
σ1ampl q ampl
ξ = av = (1.14)
p pav
• The strain accumulation and the direction of strain accumulation (cyclic flow rule)
are denoted respectively as:
q
εacc = kεacc k = (εacc 2 acc 2
1 ) + 2(ε3 ) (1.15)
ω = εacc acc
v /εq (1.16)
where εacc acc

v and εq are the cumulative volumetric and deviatoric strain components respec-
tively.
In attempt to understand the high-cycle accumulation deformation, principal remarks drawn
from this work are the following:
1.6.1 Influence of the strain amplitude
The effect of the strain amplitude εampl on the accumulation rate showed that number of
cycles caused a decrease of strain amplitude, εampl . This decrease can obviously be ob-
served within the first 100 cycles for larger amplitudes. The accumulated strain increased
proportionally with the logarithm of number of cycles. In addition the higher amplitudes
of cycles caused larger accumulation rates εacc (Figure 1.32). At higher values of the strain
amplitude εampl , the ratio of the cumulative volumetric strain to the cumulative deviatoric
strain,ω = εacc acc
v /εq , is independent of ε
ampl . For small values of strain amplitude, the ac-
cumulated volumetric strain and deviatoric strain are small which results in the inaccuracy
of ω . This suggests that the direction of accumulation is independent of the amplitude of
the deformation (Figure 1.33).
28
(a) (b)
Figure 1.32: (a) Strain amplitude; (b) Accumulation of strain as a function of number of
cycles N , Wichtmann [2005]
(a) (b)
Figure 1.33: Direction of strain accumulation in tests with different amplitudes of cycles,
εv > 0 (contraction); (a) εacc
q - εacc
p strain paths; (b) ratio ω = εacc acc as a function of
v /εq
strain amplitude ampl, Wichtmann [2005]
1.6.2 Influence of initial density, ID0
The behaviour of soils under any loading conditions is strongly influenced by many factors.
It is obvious that the initial density is one of the most important factors on strain accu-
mulation. The results obtained from variation of initial void ratio showed that the strain
accumulation increased as a function of void ratio. This is due to the decrease in stiffness
which can be attributed to an increase in void ratio. The direction of accumulation (ω) was
also found to be independent of the void ratio as shown in Figure 1.34
29
(a) (b)
Figure 1.34: (a) Strain accumulation εacc as a function of N in tests with different densities,
εv > 0 (contraction); (b) εacc acc strain paths, Wichtmann [2005]
p - εq
1.6.3 Influence of of loading frequency
Whatever load frequencies in the low range were used, there was no variation on strain
accumulation. It can be said that the loading frequency did not influence the strain accu-
mulation as well as the direction of accumulation as shown in Figure 1.35.
(a) (b)
Figure 1.35: (a) Strain accumulation εacc as a function of loading frequency, εv > 0 (con-
traction); (b) εacc acc strain paths, Wichtmann [2005]
p - εq
1.6.4 Influence of of the average stress
In case of constant stress ratio (η av ) as well as the constant amplitude of stress ratio ξ =
q ampl /pav but different mean pressures (pav ), the strain amplitude slightly increased as a
30
function of mean pressures. The strain accumulation (εacc ) significantly increased with
decrease in average mean pressure. With low average mean pressures, the dependence on
pressure was significant for large number of cycles (N) as shown in Figure 1.36. Considering
the direction of strain accumulation, however, effect of mean pressure could not be observed.
Keeping pav constant while varying η av , the strain amplitude (εampl ) slightly decreased
with increasing pav whereas the strain accumulation (εacc ) increased as a function of η av
(Figure 1.37). The stress ratio (η av ) was obviously found to influence the direction of strain
accumulation. The strain ratio (ω) obviously decreased with the variation of (pav ), the
curve showed that above a critical value of (η c ), so called characteristic threshold (η car ),
the behaviour of the material passed from the contraction to dilation phase. This was in
accordance with that performed by Luong [1980]. The characteristic threshold (CT) was
proposed by Luong in such a way that stress state below this line leads to contractive phase
or that stress state above this line leads to dilative phase (Figure1.38). This line was also
shown to be independent of pav , stress amplitude and the density.
(a) (b)
Figure 1.36: (a) Strain accumulation εacc as a function of average mean stress pav , εv > 0
(contraction); (b) εacc acc strain paths, Wichtmann [2005]
p - εq
31
(a) (b)
Figure 1.37: (a) Strain accumulation εacc as a function of of N in tests with different η av ,
εv > 0 (contraction); (b) εacc acc strain paths, Wichtmann [2005]
p - εq
Figure 1.38: Compression and dilation as a result of average stress under cyclic loading,
Luong [1980]; Wichtmann [2005]
32
In case of changing average stress q ampl between consecutive packages of cycles, this can
represent a random cyclic loading (variation of cyclic amplitudes). Generally, it is difficult
to calculate the strain accumulation due to a random cyclic loading. Then, the packages
of cycles containing a constant amplitide are replaced instead (as commonly found in prac-
tice, e.g., wind or wave loading acting on off-shore foundation, etc.). In the experimenal
observation carried out by Wichtmann [2005] and Wichtmann et al. [2010], several series of
four consecutive packages, each with 25,000 cycles, were performed. The deviatoric stress
amplitudes q ampl = 20, 40, 60 and 80 kPa were tested in different sequences whereas the
average stress pav = 200 kPa and stress ratio η av = q ampl /pav = 0.75 were kept constant.
The total strain accumulation turned out to be independent of the sequence of these pack-
ages application. The direction of strain accumulation (ω) was hardly influenced by the
consecutive packages as shown in Figure1.39. However, if the significant changes of average
stress packages especially q ampl closed to CT line were applied, these would lead to different
results. Further investigations are then necessary to overcome this ambiguity (Wichtmann
et al. [2010]).
(a) (b)
Figure 1.39: (a) Strain accumulation εacc as a function of N with changing the consecutive
packages of cycles, εv > 0 (contraction); (b) εacc acc strain paths, Wichtmann [2005]
p - εq
33
1.7 Constitutive models
1.7.1 Introduction
From experimental investigations as mentioned in previous section, it has turned out that
the interface subjected to whether monotonic or cyclic shearing load experienced a com-
pletely changed structure leading to hardening or degradation (softening). Interfaces enable
either localized dilation or contraction behaviour to occur (depending on several factors e.g.
the initial density, the local stress level and surface roughness). Various types of consti-
tutive law have continuously been proposed to describe the interface behaviour. One of
them, the simplest law used to describe the frictional contact between two solids is a rigid
perfectly plastic Coulomb condition. However, this law is not appropriate for granular soils
because the constraint on contact stress is essentially considered whereas the interface zone
is only strained in the tangential direction. Desai et al. [1985] and Desai & Nagaraj [1988]
proposed the law taking interface zone as a thin continuum into account. However, this law
may cause some problems in which the thickness of the interface should be specified. Even
though the interface thickness is commonly related to the grain size of soils, it would still
be ambiguous to specify the exact thickness of interface; moreover, the interface thickness
is significantly small in comparison with the size of the boundary.
Boulon [1989] then proposed the law in which the interface zone is considered as a two
dimensional continuum. This law does not require the constitutive parameter of interface
thickness because it is directly embodied in the constitutive equation chosen to model the
behavior of interface. It can be said that the interface is considered as a zone of zero
thickness. Subsequently, Boulon & Nova [1990] compared an incremental model with an
elastoplastic model based on general model initially proposed for sand, Nova & Wood [1979],
and successfully conformed to interfaces by changing stress and strain variables.
Appropriate and accurate constitutive models have so far been developed for characterizing
the mechanical response of interfaces which are of importance for reliable solution of the
associated boundary problems. The constitutive laws governing the mechanical interfaces
include various phenomena which can be classified according to Coulomb friction, finite fric-
tion, interface elastoplasticity, strain softening and hardening interface, and time dependent
in terms of viscoplasticity and creep.
The objective of this section is to describe the concepts of constitutive laws and how they
may be formulated for the use in interface analysis. This section begins by considering
the variety of constitutive models that have been, and still are, used to represent interface
behavior. The remainder of the section is then devoted to the extended formulations based
on, respectively, the elastoplastic and viscoplastic theories.
34
V n ½
Vn = N/A stress vector V ® ¾
¯W ¿
W = T/A >u @½
¯>w@¿
relative displacement vector u ® ¾
A = active interface area

K = prescribed normal stiffness
Figure 1.40: Basic variable involved in an interface test, Mortara et al. [2002]
1.7.2 Interface constitutive models
Considering general approach of interface constitutive models in 2-D case, the basic state
variables taken into account are then vector variables:
N T
σn = , τ= (1.17)
A A
where N and T are respectively the normal and tangential forces acting on the interface with
an area A. The kinematic state variables are the normal, [u], and tangential, [w], relative
displacements. The normal stress is considered as positive in compression and dilatancy
(increase in volume) is considered as positive (Figure 1.40).
Referring to Boulon & Nova [1990] and Boulon et al. [1995], these vectors variables can be
expressed as follows:
t = {τ σn }T (1.18)
[u] = {[w] [u]}T (1.19)
The time derivatives of these state variables are then expressed:
ṫ = {τ̇ σ˙n }T (1.20)
35
[u̇] = {[ẇ] [u̇]}T (1.21)
For any possible state of tangent loading path and the relative displacement being supposed
to be small, the constitutive equation is then given by the following relation:
ṫ = d (loading state) [u̇] (1.22)
The matrix d containing the stiffness of interface is expressed by:

ks ksn
d= (1.23)
kns kn
The diagonal components ks and kn are respectively the ”internal” shear and normal stiff-
ness. The off diagonal components ksn and kns express the coupling between shear and
normal phenomena. This coupling is activated by the dilative behaviour occurring on the
material with high relative density. However, in case the interface model is considered
in terms of uncoupled behaviour of the interface in the normal and tangential directions,
the component ksn and kns are then supposed to be omitted, consequently, the dilative
behaviour cannot be described.
In general the basic features of interface constitutive model in terms of incremental or tan-
gent path can be expressed, for instance, as;
• Interface shear test at constant σn , (CNL);

prescribed path: [ẇ] = w˙0 ; σ˙n = 0
h i
response deduced from the constitutive equation: [u̇] = − kkns ksn ·kns

n
· w˙ 0 , τ̇ = ks − kn [ẇ]
• Interface shear test at constant volume (CV);

prescribed path: [ẇ] = w˙0 , [u̇] = 0

response deduced from the constitutive equation: τ̇ = ks · w˙0 , σ˙n = kns · w˙0
• Interface shear test at constant normal stiffness ke , e ⇔ external, (CNS);

prescribed path: [ẇ] = w˙0 , σ[u̇]
˙
n
= ke
h i
response deduced from the constitutive equation: τ̇ = kek−k ksn ·kns

s
n
w˙ 0 , τ̇ = kns + ke −kn ·

w˙0
In terms of interface modeling approach, it can be deduced that the elastic perfectly plastic
law is primarily used to describe the interface mechanism. The constitutive relations and
36
associated quantities are now briefly reviewed. As the simplicity of this framework, then
certain functions such as yield (or loading) function and flow rule have to be specified.
A yield function with a non-associated flow rule is governed by Mohr-Coulomb function:
f = τ − σn tan δ (1.24)
g = τ − σn tan ψ (1.25)
The yield function (f ) is defined by the frictional angle between the soil and structure (δ)
without considering the cohesion of soils. The plastic potential (g) in which the flow rule is
allowed to use the interface dilatancy angle (ψ), can be used to reproduce the volumetric
behaviour during interface tests.
The constitutive matrix of interface behaviour can then be described by:
2
τ̇ 1 ks (1 − α) + ks kn Γ αks kn tan δ [ẇ]
= (1.26)
σ˙n ks + kn · Γ αks kn tan ψ ks kn kn2 (1 − α) [u̇]
where Γ = tan δ tan ψ . From this relation, if α = 0 then this matrix provides the elastic
behaviour, or α = 1 the plastic behaviour will occur instead. However, some shortcomings
can be found in this simple model. For instance, the first one is the appearance of tension
in the case of cohesive materials. The occurrence of tensile stress with the magnitude of
a.tanδ can be handled by the set of extra tension cut-off yield surface (Figure 1.41b) or by
introducing a new yield surface (Figure 1.41c). By using a curved yield surface (f = 0),
then a non-linear plastic potential function has to be described. The second one is that the
gradual dilation still exists during shearing load. Clearly, from experimental investigations
the dilation will vanish corresponding to the critical void ratio as shown in Figure 1.42. The
dilatancy cut-off criterion is consequently required to describe realistic feature.
W W W
tanG
a
Vn Vn Vn
(a) (b) (c)
Figure 1.41: Coulomb type yield surface extension , Boulon et al. [1995]
37
In case of ψ = 0, the elastoplastic behaviour of the interface (1.26) comes out irrelevant to
[ẇ] and the constitutive matrix can then be given as follows:
1 ks2 (1 − α) αks kn tan δ

τ̇ [ẇ] ks (1 − α) αkn tan δ [ẇ]
= =
σ˙n ks 0 ks kn [u̇] 0 kn [u̇]
(1.27)
However, this linear equation within the concept of perfect plasticity mentioned above may
be in lack of some laws describing the reasonable mechanism of interfaces. Consequently,
hardening and softening plasticity have to be required in order to provide the full set of
interface mechanism.
W [u]
reality
ks a+Vn.tanG
[w] [w]
(a) (b)
Figure 1.42: Performance of Mohr-Coulomb model in test with constant normal stress,
Boulon et al. [1995]
1.7.3 Incremental models (Implicit in the sense of Wichtmann(2005))
As mentioned in the experimental interface section, Boulon [1989] proposed the basic con-
cept for describing a general direct shear paths which can govern the set of possible responses
of the interface in case of two-dimensional condition. The basic feature of this constitutive
equation is based on the non linear relation between the incremental loadings:

[ẇ]
[u̇] dt = dt (1.28)
[u̇]
and the corresponding incremental response:

τ̇
σ̇dt = dt (1.29)
σ˙n
38
where dt is an infinitesimal time step, the set of incremental loadings and responses nor-
malized from the initial state can then be expressed as:
1/2
k[u̇]k dt = [ẇ]2 + [u̇]2 dt (1.30)
Supposing (λ ,µ) is an arbitrary space in which the possible incremental loading paths can
be described as the points of the unit circle. The corresponding points of the incremental
response space (ξ , η) depend on two parameters (λ and µ) connected by one relation (λ2
+ µ2 = 1). The direction of incremental loading and the corresponding response are then
defined as follows:

λ 1 [ẇ]
= · (1.31)
µ k[u̇]k [u̇]

ξ 1 τ̇
= · (1.32)
η k[u̇]k σ̇
The interface material is also supposed to exhibit the continuous incremental behaviour
for loading direction. From the initial state, the directional incremental parameters λ and
µ enable the set of possible loading and the response to be mapped in the corresponding
incremental spaces. Then the approximation of possible incremental responses can be set
up by interpolation, just knowing a small number N of basic incremental responses:
N
ξ X ξi τ̇ τ̇ ([ẇ] , [u̇] , initial state)
= Wi −→ = (1.33)
η ηi σ̇ σ̇([ẇ] , [u̇] , initial state)
i=1
where Wi are interpolation functions (e.g., polynomial, trigonometric).

The incremental paths based on three different conditions (e.g., constant normal load (CNL),
constant volume (CV) and constant normal stiffness (CNS)) are then obtained by derivation
of analytical expressions in which, for instance, τ and [u] can be represented as a function
of [w] (e.g., τ ([w]), [u]([w]), etc.) in case of CNL tests. Figure1.43 illustrates typical key
parameters for describing the basic direct shear tests.
In case of CNL test, considering the phase of [w] ≤ [w]CN L+ and [w] ≥ [w]CN L+ , as shown
in Figure 1.43(a) and (b), where [w]CN L+ is the tangential displacement corresponding
to the ultimate state (τ /σn0 )+ , [w]p and [u]p are respectively the tangential and normal
displacements at peak stress ratio, the parameters can be given as follows:
39
§ W ·
¨ ¸ >u @
§ W · ¨© V n 0 ¸¹ w>u @
¨¨ ¸¸ >u @CNL w>w@f
© V n0 ¹ p
§ W ·
¨¨ ¸¸
© V n0 ¹ >u @ p
>w@
§W · >w@ p >w@CNL
¨¨ ¸¸
© V n ¹CNL 0 >w@ >u @
>w@ p >w@CNL >w@0
(a) (b)
§W · Vn
¨¨ ¸¸
§W · ©V n ¹ § § W ··
¨¨ ¸¸ ¨ w¨¨ ¸¸ ¸ V n § wV n ·
© V n ¹ ¨ ©V n ¹ ¸ ¨ ¸
¨ w>w@ ¸ © w>w@ ¹f
¨¨ ¸¸
© ¹f
V n0
§ V n V n0 ·
§W · ¨ ¸
© w>w@ ¹0
¨¨ V ¸¸
© n ¹CV 0 >w@ >w@
>w@CV >w@CV
(c) (d)
Figure 1.43: Key parameters for description of the basic direct shear tests; (a), (b) constant
normal load (CNL); (c), (d) constant volume test (CV), Boulon & Nova [1990]
• phase of [w] ≤ [w]CN L+

" #
τ τ τ τ
= fCN L , [w]p , , [w]CN L+ , (1.34)
σn0 σn CN L0 σn0 p σn0 +

[u]
[u] − [u]0 = gCN L , [w]p , [u]p , [w]CN L+ , [u]CN L+ (1.35)
[w] 0
• phase of [w] > [w]CN L+

τ τ
= (1.36)
σn0 σn +

∂[u]
[u] − [u]+ = · ([w] − [w]CN L+ ) (1.37)
∂[w] ∞
Similarly, in the case of CV or CNS (Figure 1.43(c) and (d)) in which the normal stress
varies during shearing load, where [w]CV + is tangential displacement corresponding to the
40
peak state (τ /σn0 )+ , the parameters can also be given as follows:
• phase of [w] ≤ [w]CV +

τ τ τ
= fCV , [w]CV + , (1.38)
σn0 σn CV 0 σn +

σn − σn0
σn − σn0 = gCV , [w]CV + , σn+ (1.39)
[w] 0
• phase of [w] > [w]CV +
 
τ

τ
∂ στn
− =   · ([w] − [w]CV + ) (1.40)
σ σn + ∂[w]
∞

∂σn
σn − σn+ = · ([w] − [w]CV + ) (1.41)
∂[w] ∞
It can be noted that for defining the entire set of incremental responses, unloading incre-
mental paths are then assumed to have the initial slopes ([w] = 0) of the corresponding
loading incremental paths. In the final manner, the constitutive equation can formally be
written via Euler’s theorem for homogeneous functions:
" #
∂ τ̇ ∂ τ̇
τ̇ ∂[ẇ] ∂[u̇] [ẇ]
= ∂ σ˙n ∂ σ˙n (1.42)
σ˙n ∂[ẇ] ∂[u̇]
[u̇]
1.7.4 Elastoplastic models (Implicit in the sense of Wichtmann(2005))
The majority of interface models have so far been devoted to elastoplastic framework in-
corporating whether hardening or softening law. These models are based on classical soil
mechanics and are able to reproduce the prominent behaviour of interfaces. The plastic
function of the model is now taken to be a generalization of yield function as a hardening
parameter (Aubry et al. [1990] ; Mortara [2001] ; Gennaro & Frank [2002] ; Mortara et al.
[2002] ; Shahrour & Rezaie [1997] ; Boulon et al. [2003] ; D’Aguiar et al. [2008] ; D’Aguiar
et al. [2011]).
Following Mortara [2001] , Mortara et al. [2002] , Boulon et al. [2003], the normalized
displacement [wn ] is defined in incremental form as;
Z
[wn ] = [ẇn ] dt (1.43)
t
41
where [ẇn ] is the ratio between the plastic tangential displacement rate [ẇp ] and the plastic
tangential displacement at failure [wp ] corresponding to the peak value of the stress ratio
ηp .
[ẇp ]
[ẇn ] = (1.44)
[wp ]
From the definition of normalized displacement, the locus of the stresses at failure corre-
sponds to the plastic function for [wn ] = 1. The plastic functions of the model are then
given as:
√
f= τ 2 − ασn = 0 (1.45)
where α is a hardening-softening function which defines the rule of expansion and contrac-
tion of the plastic function. Function α then represents the slope of the stress envelopes for
the value of [wn ] = 1 (Figure 1.44).
shear stress W (kPa)
Experimental data at failure

Failure line
normal stress Vn (kPa)
Figure 1.44: Exprimental data at failure and plastic function for [wn ] = 1, interface test on
Toyoura (silica) sand (Dr = 85%) with roughness Rmax = 60 µm, Mortara et al. [2002] ,
Boulon et al. [2003]
Figure 1.45 also shows the α functions in dependence on [wn ] utilized to interpolate αexp
values. The function α([wn ] is given by the following equation:
h i
α ([wn ]) = αc (ω[wn ] + 1)ψ − 2 exp −L[wn ]M + αc

(1.46)
42
Dp
D functions Dc
normalised displacement [wn]
Figure 1.45: Exprimental and model a values, interface test on Toyoura (silica) sand (Dr =
85%) with roughness Rmax = 60 µm, Mortara et al. [2002] , Boulon et al. [2003]
where αc is the asymptotic value of the function α([wn ]), ω and ψ are curve fitting param-
eters. By defining the conditions in α([wn ]) as:
dα
α|[wn ]=1 = αp , | =0 (1.47)
d[wn ] [wn ]=1
where αp is the peak value of α([wn ]) function as shown in Figure1.45. For L and M , these
can now be generated:
i
αc h ψ
L = ln (ω + 1) − 2 (1.48)
αp − αc

αc ωψ
M= (ω + 1)ψ−1 (1.49)
αp − αc L exp {L}
According to the classical theory of plasticity, the derivatives of plastic potential g surface
provide the relative magnitudes of plastic displacement rates:
∂g ∂g
[ẇp ] = λ̇ , [u̇p ] = λ̇ (1.50)
∂τ ∂σn
where [ẇp ] and [u̇p ] are respectively the plastic rate of normal and tangential displacements
and λ̇ is a scalar plastic multiplier. Based on the orthogonality of the plastic incremental
displacement vector to the plastic potential, it is possible to examine the plastic displace-
ment by considering the variation of dilatancy with stress ratio. The dilatancy (d) can
therefore be written as:
43
Figure 1.46: Flow rule of an interface test on Toyoura (silica) sand (Dr = 85%) with
roughness Rmax = 60 µm, Mortara et al. [2002] , Boulon et al. [2003]
[u̇p ] dτ
d=− p
, − (1.51)
[ẇ ] dσn
Rewriting this relation in term of η and d gives:
dσn dη
= (1.52)
σn d+η
Figure 1.46 shows the typical stress-dilatancy relation for an interface test. The flow rule
observed can be considered as a bilinear relation which can be described for hardening and
softening conditions, respectively, as:
η = a1 d + b1 for [wn ] ≤ 1 (1.53)
η = a2 d + b2 for [wn ] ≥ 1 (1.54)
where a1 , b1 , a2 and b2 are respectively the slopes and the interceptions of the pre-peak
and post-peak flow rule in the stress-dilatancy plane. The convergence of flow rules at the
peak of direct shear test, [wn ] =1, is then given as:
ηp = a1 dmax + ρηc = a2 dmax + ρηc (1.55)
where ηp and ηc are respectively the stress ratio for peak and ultimate conditions, and
dmax is the maximum dilatancy. The parameter ρ is the ratio between the stress ratios
corresponding to the condition d = 0 in hardening and softening conditions. The expression
of plastic potential which satisfies the required conditions is:
44
hardening
softening
shear stress W
Kp
1
normal stress Vn
Figure 1.47: Plastic potential for failure conditions; interface test on Toyoura (silica) sand
(Dr = 85%) with roughness Rmax = 60 µm, Mortara et al. [2002] , Boulon et al. [2003]
" −(1+a)/a #
√ b σn
g = τ2 − σn 1 + a = 0 (1.56)
1+a σc
where a and b are the generic slope and the interception of the flow rule in the η −d diagram
and σc is the normal stress corresponding to d = 0. This plastic potential is similar to that
of Nova & Wood [1979] and is shown in Figure 1.47 for failure conditions.
In elastic domain, the shear stiffness is proportional to the normal stiffness with a constant,
CK :
kse = CK kne (1.57)
It can be noted that the elastic constant does not influence the response provided the
modeling is satisfactory by having an accurate order of magnitude.
For monotonic interface shear test, the simple elastoplastic or incremental models can be
used to simulate the interface mechanism with sufficient accuracy. However, for more com-
plex loading programs involving cyclic loading, more complex rules should be examined in
order to describe more realistically the effects occurring for these loading conditions. The
applicability of the model can be extended by formulating a cyclic surface (f0 ) subjected
to kinematic rotational hardening inside the isotropic plastic surface (absolute limit sur-
face, f ). Cyclic surface which separates domains of active loading and elastic unloading is
described by
√
f0 = τ 2 − α0 σ n = 0 (1.58)
45
Where τ and σ n are the stresses in a reference system with the surface (1.58) rotated by an
angle θ with respect to axis τ = 0, Figure 1.48;
2 arctan D
shear stress W
stress state
normal stress Vn
Figure 1.48: Isotropic and kinematic surfaces of the model, Mortara et al. [2002] , Boulon
et al. [2003]
σ n = σn cos θ + τ sin θ (1.59)
τ = −σn sin θ + τ cos θ (1.60)
where α0 is hardening-softening function cyclic mechanism which specifies the cyclic mech-
anism, and is supposed to be a constant proportion of the maximum amplitude of the
isotropic surface:
arctan α0 = Cα arctan αp (1.61)
where Cα is a parameter of the model which has to be interpreted numerically. The angular
distance θr between the isotropic plastic surface (f ) and cyclic surface (f0 ) as shown in
Figure 1.48 is given by:

∂f0 τ
θr = arctan α − sgn arctan (1.62)
∂τ σn
The hardening modulus for plastic state within the isotropic plastic surface is given by an
interpolation function:
θr
H = Hα + h (1.63)
θr0 − θr
46
where Hα is the hardening modulus relative to the isotropic mechanism and the angle θr0
is the value of θr corresponding to the reversal loading. Function h which describes the
dependency of hardening modulus on shear stress mobilization and of irreversible normal
displacement and relative position of two surfaces is given by:
[upmax ]

αp σ n δ nθr0
h= γ[wn ] Rαθ 1 + χh (1.64)
[wp ] s0
where γ , δ , n and χh are parameters of the model while s0 is an arbitrary reference

displacement. The quantity Rαθ is given by:
2(arctan α − arctan α0 )
Rαθ = (1.65)
θr0
With the condition Rαθ 6= 1, this means that the reloading proceeds without reaching
isotropic plastic surface (f ).
To indicate how the above concepts of cyclic interface behaviour work, Figure 1.49 can
clearly illustrate the typical paths of cyclic CNL test in τ − σn and τ − [w] planes. Initially,
the two domains move solidly on first loading along path AB and the interface response
is governed by the external surface. Once the first reverse loading occurs, the behaviour
is elastic on initial unloading (AB) and is sufficient to invoke the cyclic surface. Along
BC the interface behaves elastoplastically and the hardening modulus is governed by the
equation in (1.63). The next reverse of shear loading at C produces an elastic response.
With further loading the interface behaviour is still defined by the cyclic domain until the
stress state reaches E which coincides with A on the isotropic domain. Further loading EF
will eventually invoke the isotropic domain.
The interface subjected to cyclic loading usually exhibits a progressive densification which
leads to an important role in determining the cyclic degradation of shear resistance under
CNS tests. For increasing number of cycles, the cyclic flow rule undergoes a progressive
variation that can be schematized as a parallel translation in comparison with that of static
condition. The following form of η = a1 d + b1 is then proposed for cyclic condition:
[upmax ]
η = ad + b + ∆b = ad + b + χb (1.66)
s0
where χb is a parameter of the model. However, the direction of incremental displacement

which is executed from isotropic potential (Pastor et al. [1985]) has to be defined when a
loading reversal takes place. During cyclic loading the following definition is given as:

∂f0 ∂g 1 for Loading (L)
ϕ = sgn = (1.67)
∂τ ∂τ −1 for Unloading(U)
47
Figure 1.49: Behaviour of the cyclic model under CNL condition , Mortara et al. [2002]
which can govern whether hardening or softening conditions. With gradients of plastic
potential and plastic functions:
" # " # " # " #

∂f ∂f0 ∂g ∂g
∂σn ∂σn ∂σn ∂σn
n= ∂f , n0 = ∂f0 , vL = ∂g , vU = ∂g
, (1.68)
∂τ ∂τ ∂τ − ∂τ
the elastoplastic matrices of the isotropic plastic surface (D) and cyclic surface (D0 ) are
then given by the following relations:
D e v L nT D e
e e De vL/U nT0 De
D=D − , D0 = D − (1.69)
Hα + nT D e v L H + nT0 De vL/U
where vL/U = vL for ϕ = 1 and vL/U = vU for ϕ = -1.
1.7.5 Pseudo-creep models (Explicit in the sense of Wichtmann(2005))
Appropriate model describing the interface mechanism is a major object in engineering

applications. One of the considerable conditions is the time rate effect in which the conven-
tional theory of plasticity cannot describe the interface mechanism adequately. This time
rate effect is closely related to a degradation of properties which can be revealed by creep
and relaxation phenomena and by the influence of strain rates. Furthermore, the interface
48
behaviour can significantly be crucial when time-dependence is taken into account. A fur-
ther condition in which time rates plays an important role is that when the interface is
subjected to dynamic transient loading. For instance, interfaces subjected to cyclic loading,
even with small amplitude, may lead to a significant long term plastic displacement. As
an experimental observation the yield function of material under high strain rates can be
greater than that under the low (quasi-static) value.
However, interface investigations have rarely taken time-dependence into account. In order
to define the interface mechanism realistically, the theory of viscoplasticity providing a
unified approach to the problem of creep and plasticity has to be proposed. In general, a
widely-used viscoplastic formulation is the Perzyna model, Perzyna [1966].
The major concept of this model, so called overstress, is that the yield surface does not have
a limit on the possible stress states and then viscous effects related to the stress component
can exceed this yield surface. Figure 1.50 describes the definition of the overstress which is
defined as the distance in stress space between the current stress state P and the static yield
surface fs . The dynamic loading surface fd on which the current stress state P is located
depends on the stress state and viscoplastic work. The direction of viscoplastic strain
rate is perpendicular to the plastic potential surface g. In Perzyna’s overstress theory,
viscous effects are negligible in elastic domain, e.g. no viscous strains take place within
the static yield surface which corresponds to the classical yield surface associated with
time-independent plasticity. So far, the characteristics as well as numerical formulations of
Perzyna model have been devoted by many authors, e.g. Owen & Hinton [1980]; Flavigny
& Nova [1987]; Samtani et al. [1996]; Heeres et al. [2002]; Hicher & Shao [2002]; Liingaard
et al. [2004]; Yin [2006]; Stijn et al. [2007]; Yin & Karstunen [2008]; Karstunen et al. [2008].
viscoplastic H ijvp
regime
non-viscous overstress g
regime
fs fd
V’ij
Figure 1.50: Overstress F is defined as the distance between current stress state P on the
dynamic yield surface fd and the static yield surface fs , Liingaard et al. [2004]
49
In view of elasto-viscoplastic model of soil-structure interface, the proposed constitutive
model requires the simplest possible formulation which can qualitatively reproduce the
entire feature of experimental observation. Referring to Samtani et al. [1996] and Armand et
al. [1998], the displacement vector can be additively decomposed into elastic and viscoplastic
components:
[u̇e ] [u̇vp ]

[u̇]
s= = + (1.70)
[ẇ] [ẇe ] [ẇvp ]
where the superimposed dot denotes the time derivative. Additionally, if the shear stress
and normal responses are taken to be uncoupled, the elastic part of velocity vector is then
related to stress rate vector as follows:
[u̇e ]

k 0 τ̇
= s (1.71)
[ẇe ] 0 kn σ˙n
where ks and kn are the elastic normal and shear stiffness of the interface which are expressed
as follows:
dσn dτ
kn = , ks = (1.72)
d[u] d[w]
By following Perzyna’s original approach, the viscoplastic velocity vector is defined by a

viscoplastic flow rule of the formulation:
∂g
ṡvp = γΦ (F ) (1.73)
∂σ
In this formulation the viscoplastic potential g gives the direction of ṡvp , the viscous nucleus
Φ is a scalar function of the yield function F and γ is a material constant which is known
as the fluidity parameter with the dimension of inverse time. Adopting anisotropic (for
forward and reversal shear loading) Mohr-Coulomb plasticity model, the yield surface F
(Figure 1.51), according to Armand et al. [1998] and Papadopoulos et al. [1998], is given as:
F1,2 = τ + (r ± k) σn (1.74)
with the definition of
1
r= (µ1 − µ2 ) (1.75)
2
1
k= (µ1 + µ2 ) (1.76)
2
50
W
r+k=P1 F1=0
forward
r=1/2 (P1-P2) f=r
Vn
r-k=P2 reverse
F2=0
Figure 1.51: Yield surface representation, Armand et al. [1998]
where µ1 and µ2 are friction coefficients (µ = tan δ) in forward and reversal direction in a
parallel manner. The plastic potential in which the dilatancy angle (ψ) is described is then
given as:
g1,2 = τ + tan (ψ1,2 ) σn + const (1.77)
The viscous nucleus Φ(F ) describes the evolution of the viscoplastic strain vector. By using
the Macauley bracket h i, a simple representation of viscous nucleus is then given as:

F −F/σ if F > 0
Φ (F ) = = (1.78)
−σn 0 if F ≤ 0
or it can be rewritten as:
F

f + µ1,2 if f1,2 ≥ 0 with f =
Φ (F ) = −σ (1.79)
0 else
The definition of viscous nucleus Φ(F ) can implicitly be described that the viscoplastic
velocity is not zero even if the current state of stress lies outside the considering yield
surface. The material constant introduces the physical time dimension into the problem.
Considering for the limit γ → ∞, the constitutive model turns out to be plastic and its
time dependency then disappears.
Following the element tests on high-cycles accumulation, several studies on the cyclic be-
haviour of sand concerning a large number of cycles (N > 103 ) were carried out in the
approach to predict the accumulated deformations (pseudo-creep). In general, there are
two main approaches for describing the accumulation due to cyclic loading. The most
51
conventional one, so-called implicit (elastoplastic) model, is capable of describing the accu-
mulation for N < 102 . For instance, the numerical error which normally occurs in modeling
would increase significantly during an application of a large number of cycles. This denotes
the importance of the explicit model in which the number of cycles N is treated instead of
time.
Niemunis et al. [2005] and Wichtmann et al. [2005] proposed the rate of strain accumulation
(Dacc ) based on the cyclic triaxial tests as:
Dacc = mfampl f˙N fp fY fe f π (1.80)
with the following functions which describe the influence on the rate of strain accumulation;
• fampl : amplitude, shape and polarization of the strain loop, summerized in the scalar
εampl
• fp : average mean stress pav
av
• fY : average stress ratio Y
• fe : average void ratio e
• f˙N : number of cycles N, if εampl = constant.
• fπ : polarization changes
m is the unit tensor expressing the direction of accumulation(flow rule) which is only de-
pendent of average stress ratio (η av ). Table1.1 shows the definition of these functions except
fπ .
Table 1.1: Summary of the partial functions fi for triaxial tests performed by Niemunis et
al. [2005]
Function Mat. Constants
2
fampl = εampl /εampl
ref εampl
ref
˙ CN 1 CN 2
fN = 1+CN 2 N + CN 1 CN 3 CN 1 , CN 2 , CN 3
h av i
fp = exp CP ppref − 1 CP , pref
av
fY = exp CY Y CY
(Ce −e)2 1+eref
fe = 1+e (Ce −eref )2 Ce , eref
fπ = 1 if polarization of the strain loop = constant
Messast et al. [2006] formulated a simple model describing the volumetric strain accumula-
tion (εcvN ). This formulation is based on the cyclic parameters of the first cycle:
52
N
εcvN = 1
(1.81)
εcv1 + Nεc−1
v∞
where εcv1 and εcv∞ are the strain accumulation when N = 1 and N → ∞, respectively. The
volumetric strain accumulation at N → ∞, εcv∞ , can be expressed as a function of ∆η and
η av :

η av
η car −1
εcv∞ = ε∞
v0 av
(1.82)
η
1− η peak
As can be seen in Figure 1.52, the function can then be written in the following form:
C1 ∆η
ε∞
v0 = (1.83)
∆η + C2
where C1 and C2 are the coefficients depending on η av and the void ratio.
Figure 1.52: Strain accumulation at infinity εcv∞ as a function of ∆η and ∆av , Messast et
al. [2006]
1.8 Conclusions
The behaviour of interfaces has widely been the subject of experimental studies. The
majority of earlier studies dealt with the monotonic behaviour of interfaces. The behaviour
of interfaces was found to depend on the soil properties and surface roughness. Subsequently,
experimental studies on cyclic behaviour of interfaces were performed to understand the
53
mechanics of the interaction. From these studies, it is possible to conclude that the cyclic
interface behaviour is influenced by the amplitude of tangential displacement, normal stress,
density of sample, number of loading cycles, position and amplitude of the cycles within
the stress space.
The use of a constant normal stiffness to investigate the behaviour of soil-structure interfaces
is based on the observation that soil deformation is concentrated in a thin region close to
the structure. In model pile tests, cyclic loading has been found to decrease the frictional
resistance of the surrounding soil. This phenomenon has been attributed to a decrease in
the normal stress due to cumulative contraction of the soil within the interface shear zone.
It can be concluded that the reduction in normal stress associated with shear stress is not
only due to the normal stiffness, but also due to the increasing amount of shear displacement
at the interface with an increase in number of cycles.
On high cycle accumulation the intensity of strain accumulation depends on several factors,
e.g. initial density, initial confining stress (pav ), amplitude of average stress (q ampl ) and the
stress ratio (η av ). But the direction of strain accumulation (ω) depends exclusively on the
stress ratio (η av ).
The concept of such models aiming at reproducing soil-structure interface behaviour is
formulated by using the same physical principles as those of soil mechanic models. The
interface models which have been described can provide a satisfactory and consistent for-
mulation of interface behaviour. However, the progress in modeling interface behaviour is
still required in order to achieve a more realistic overview.
54
Chapter 2
Interface direct shear tests
2.1 Introduction
The work presented in this chapter focuses on the identification and characterization of
granular soil-structure interface behaviour, and more particularly of interface behaviour
concerning a large number of cycles (typically 104 ) with small amplitude of cycles in terms
of shear stress. The purpose of this section is the presentation of an innovative direct
shear device developed at laboratory 3S-R. This device is specifically designed to access
the domains of soil-structure interfaces whether under constant normal stress or unit load
(CNL) or at constant normal stiffness (CNS) conditions (Hoteit [1990]; Moutraji [1992]).
Therefore, a large part of this chapter is devoted to the illustration of the experimental
facilities implemented during this study and the interpretation of the data delivered. Most of
interface direct shear tests carried out in this work focus on rough plate under CNL and CNS
conditions. For smooth plate, some tests are additionally presented. The results, analyzed
and processed, from the various experimental campaigns are presented and summarized in
the following sections.
2.2 Interface direct shear devices
The mechanism of interface soil-structural materials was studied in detail by several re-
searchers at 3S-R laboratory at which the tests with constant normal stress (CNL) and
constant volume (CV) are carried out by using a modified direct shear test. Hoteit [1990]
carried out a complete direct shear test to impose normal stiffness by using two types of
sand and two different values of roughness. Moutraji [1992] later performed and modified
interface ditrect shear tests on silt-steel by measuring relative displacement and stresses in
normal and tangential directions. For interface direct shear box used in this study, the lower
half-box is replaced by a plate whose roughness is variable (smooth, rough) according to
the tests and the purpose which are to simulate the side surface of the structural materials
anchored in the soil. The sample box which is of cylindrical form is placed at a calibrated
55
distance of the plate. The normal stress (σn ) is then applied vertically via a piston (top
cap), while the tangential stress (τ ) is obtained by a horizontal relative movement (tangen-
tial displacement [w]) of the plate. Contrary to more traditional tests which are carried
out only under both constant normal stress (CNL) and constant volume (CV) conditions,
the system of current loading allows the realization of various conditions, in particular the
constant normal stiffness (CNS) condition.
The direct shear apparatus used in this study is shown in Figure 2.1. The main aspect of
φ 60 mm shear box apparatus was modified to enable the normal load to be applied by a
generating engine to provide the constant normal stiffness condition. The conditions with
constant normal stress (CNL) and constant volume (CV) can be regarded as ”extreme”
conditions, framing the real paths followed by the interface during the mobilization of side
friction along a pile, according to the simplified model suggested by Boulon [1989]. This
test, direct shearing with imposed normal stiffness, is carried out by using a control between
the variation of normal stress (σn ) acting on the sample and the normal displacement ([u])
at the top part of sample box.
Figure 2.1: Direct shear apparatus used in this research
The generating engine is employed to apply the normal load on the interface. The operation
of this generating engine is controlled by a computer to impose a constant normal stress
(k = 0) or a constant stiffness (k 6= 0) conditions. It consists in controlling the variations
of the normal stress (σn ) to normal relative displacement [u] according to the relation:
56
C = ∆σn − k∆ [u] (2.1)
Indeed, to measure the values of different stress and displacement transducers in real time,
one is able to control a physical value measured at an imposed set over time. Therefore, this
allows the performance of tests with different stress paths. Possible paths of direct shear
according to value of k are:
• Case I. k = 0 ; constant normal stress
• Case II. k = ∞ ; constant volume
• Case III. k > 0 ; constant normal stiffness (k)
Figure 2.2 shows a schematic illustration of the control system. The generation of normal
stress is ensured by traditional micro engine in two directions, ensuring loading or unloading
of the sample. During shearing, the set (2.1), where k is the imposed normal stiffness, is
applied with a precision of 0.1 kPa. This is capable of following the most general path of
direct shear tests. Shearing is induced on the sample by tangential displacement of the
structural surface which can directly be controlled by the system. Using the shear rate with
low range, it is assumed that there is no influence of shear rate on test results. In this study,
the compatible maximum [ẇ] with a sufficient data acquisition is then used with a shear
displacement rate of 0.5 mm/min. In case of cyclic tests under controlled shear stress, two
thresholds (high and low, adjustable) of shear stress are prescribed, causing a reversal of
shear direction when they are reached. Electrical displacement transducers, monitored by
a computer system, are used to measure displacements and the applied loads in calibrated
proving rings.
Figure 2.2: Principle of the direct shear test to impose normal stiffness (k), Hoteit [1990]
57
The calibration of transducers based on displacement transducer (LVDT) and load cells is
a procedure which converts an electrical value into a physical quantity. This electrical value
varies according to the amount of the physical quantity which is under the measure. For
calibrating the load cells, as the load placed upon the load cell, the electrical resistance
changed due to the flexion of the cell under loading. The electrical value produced by the
load cell was picked up by a digital-analogical converter, and then sent to a computer for
displaying and recording. To compute the calibration constants, the relationship between
the change of electrical values and loads is obtained from the slope of linear regression line.
The calibration of LVDTs (resolution 0.01 mm) can be done on the same way. Figure 2.3
shows typical calibrations of these transducers.
0.5 0.5
0.4 0.4
0.3 0.3
shear force (kN)
y = 0.0923x + 0.0487 y = -0.2594x + 0.0826

0.2 R² = 0.9999 0.2
R² = 0.9998
normal force (kN)
0.1 0.1
0 0
-6 -4 -2 0 2 4 6 -3 -2 -1 0 1 2 3
-0.1 -0.1
-0.2 -0.2
-0.3 -0.3
-0.4 -0.4
-0.5 -0.5
electrical value (v) electrical value (v)
4.5 16
4 14
normal displacement (mm)
shear displacement (mm)
3.5
12
y = 8.1102x + 8.523
3 y = -2.5312x + 2.4507 R² = 0.9999
10
R² = 0.9995
2.5
8
2
6
1.5
4
1
0.5 2
0 0
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
electrical value (V) electrical value (mm)
Figure 2.3: Calibration of force and displacement transducers
During interface direct shear tests, the signal of all sensors were recorded every five seconds
by means of a data acquisition system. The measured and recorded variables are the stress
vector applied to the interface (normal, σn ( > 0 in compression) and shear, τ components
and the displacement vector on soil-structure (components normal [u] > 0 dilating) and
tangential [w]). Table 2.1 shows the relationships of state variables in data acquisition
system.
58
Table 2.1: Relationships of interface variables of direct shear test in data acquisition system
State variable Relation
τ −(∆V )0.2594/A
σn ((∆V )0.0923 + c)4.76/A
[u] 2.5312(∆V )
[w] 8.1102(∆V ) − 0.00272727τ
[v] −1.9239(∆V )
where ∆V is the change of electical values recorded from the beginning. A is a cross section
area of shear box. Taking into account the vertical movement of load cell, the coefficient c
is then described as c = 0.0026[v]. The vertical movement of shear box, [v], is additionally
measured in order to find the solution of leakage of sample which will be discussed later.
However, the LVDTs used for measuring the displacements have the range limits. For
instance, the LVDTs for measuring the normal and tangential displacements provide high
precision and accuracy in the ranges of ± 2 mm and ± 6.5 mm, respectively. In some
cases of cyclic interface shear tests, especially in CNS test series, the horizontal relative
movement of the plate would exceed the range of 13 mm but the tangential displacement
is still recorded in the range of 13 mm. This provides the partial measurement error of
tangential displacement. Installation of these sensors on the apparatus is shown in Figure
2.4.
Figure 2.4: Position of three LVDTs on direct shear apparatus
59
130
ID y 90% ; Vn = 120 kPa ; CNL test
125
Vn (kPa) 120
115
mean Vn = 118 kPa

110
105
100
0 200 400 600 800 1000 1200 1400 1600
time (s)
Figure 2.5: Typical normal stress applied to the sample, σn = 120 kPa, ID0 = 90 % under
CNL condition
140 0.2
ID y 90% ; Vn = 310 kPa ; cyclic CNL tests ID y 90% ; Vn = 310 kPa ; cyclic CNL tests
120 0.15
W (kPa)
[w] (mm)
'W = 40 kPa
100 0.1
'W = 20 kPa
'W = 10 kPa
'W = 20 kPa 'W = 10 kPa
80 0.05
'W = 40 kPa
60 0
0 200 400 600 800 1000 0 200 400 600 800 1000
time (s) time (s)
120 0.5
ID y 30% ; Vn = 310 kPa ; cyclic CNL tests ID y 30% ; Vn = 310 kPa ; cyclic CNL tests
110
0.4
100
[w] (mm)
W (kPa)
0.3
90
80
0.2 'W = 40 kPa
'W = 10 kPa 'W = 20 kPa
70
'W = 20 kPa 0.1
60
'W = 40 kPa 'W = 10 kPa
50 0
0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 1400
time (s) time (s)
Figure 2.6: Typical shear stress applied to the sample and shear displacement [w] output
on cyclic CNL tests, σn = 310 kPa with cyclic amplitude ∆τ of 10, 20 and 40 kPa
60
The typical normal stress applied to the sample during shear loading, in case of CNL
condition, from the generating engine is shown in Figure 2.5. Even though this test is
under CNL condition, there is still a variation in normal stress (typically in the range of
∆σ = ± 5 kPa). This small variation of the normal stress is due to the generating engine
in two directions. In case of cyclic tests, the cycles are carried out between two shear stress
thresholds (shear stress-controlled tests). Even though the cyclic forms are not somehow
uniform (Figure2.6), depending on the level of mean cyclic stress ratio (ηcm ) and density
index (ID0 ), these cyclic tests still provide the reasonable results on the ground that the
mean cyclic paths are principally considered.
2.3 Tested materials
2.3.1 Standard Fontainebleau sand
In several laboratories in France, standard Fontainebleau sand is commonly utilized. This

sand is siliceous sand. It has a color that ranges between grey and white. The characteris-
tics of standard Fontainebleau sand which was used in this research are shown in Table 2.2.
For Fontainebleau sand NE34, SOLCYP* Project provides the values of emin = 0.545 and
emax = 0.866, which may be fairly representative and usable in our first approach to define
three states (loose (ID0 = 40%), medium (ID0 = 65 − 70%) and dense (ID0 = 90%)). These
values correspond to maximum and minimum dry densities of 1.72 g/cm3 and 1.42 g/cm3
respectively (the specific gravity of the material G is taken equal to 2.65 g/cm3 , practically
pure silica). It is worth noting that these values should then compare to those that have
been obtained by using the standard in a rigorous way (proposed by A. Puech and J. Gar-
nier, Fugro Geotechnical laboratory), which we take as the reference. The measurements
of emin and emax on the sample of Fontainebleau performed by Fugro Geotechnical Labo-
ratory following the French standard with G = 2.70 g/cm3 are 0.50 and 0.89, respectively.
Maximum dry density is close to that found by SOLCYP Project, whereas minimum dry
density is farther. However, a good correspondence still exists between these values.
An important element of the characterization of a set of grains is the grading curve. Figure
2.7 shows two curves of particle size distribution obtained by two different methods. A
typical curve obtained by sieving method still serves as a reference curve for defining the
characteristics of the material. The other method by laser granulometry provides an esti-
mate of the average size of each particle. This method thus shows a shift of the grading curve
to the right (larger diameter). The difference between these curves is due to the sieving
method taking into account the weight of sand whereas laser granulometry method which
takes into account the volume mass of sand provides the average particle size. However,
these two methods still provide the parallel curves (the same uniformity). The mean grain
size D50 is identified to be 0.23 mm for sieving method and 0.25 mm for laser granulometry.
With the uniformity coefficient (Cu = D60 /D10 ), from sieving method Cu ≤ 4, this sand
can be considered to be poorly graded or uniform.
61
Table 2.2: Properties of standard Fontainebleau sand according to SOLCYP* Project
D50 G ρd max ρd min emax emin Cu = D60 /D10
3 3 3
(mm) (g/cm ) (g/cm ) (g/cm )
0.23 2.65 1.72 1.42 0.866 0.545 1.72
* Research on behaviour of piles subjected to cyclic loading Project (French National Pro-
gram)
100
90
80 Laser
Lasermethod
method
finer by weight (%)
Sievingmethod
sieving method
70
60
50
40
30
20
10
0
0.01 0.1 1 10
particle size (mm)
Figure 2.7: Fontainebleau sand and grain size distribution
With the minimum and maximum void ratios (emin , emin ) or minimum and maximun dry
densities (ρd min , ρd max ), the density index (ID0 ) used to classify the dense or loose sample
in this study was then given as:
emax − e0 ρd max (ρd0 − ρd min )

ID0 = = (2.2)
emax − emin ρd0 (ρd max − ρd min )
where e0 and ρd0 are respectively the void ratio and the dry density at the beginning of a
test (before applying normal stress). Within this study, two distinct densities ID0 = 90±3%
and ID0 = 30 ± 3% represent dense (γd0 ≈ 16.55 kN/m3 ) and loose (γd0 ≈ 14.74 kN/m3 )
samples respectively.
2.3.2 plates
Having mentioned in the previous chapter, the surface roughness of structural materials
were quantified as the modified roughness in term of a maximum height Rmax , L = 0.20
mm (Uesugi & Kishida [1986]), which is the relative height between the highest peak and the
lowest valley along a surface profile over the gauge length L = 0.20 mm. In this study, the
62
rough plates mainly used to describe the interface behaviour were made by gluing on a steel
plate 1 cm thick layer of 0.2 mm from a mixture of araldite (epoxy) and Fontainebleau sand
(30 g of araldite per 100 g of sand passing the sieve with mesh size of 630 µm and retained
on the sieve with mesh size of 315 µm), this method then represents a high roughness
(Figure2.8). The value of roughness (Rmax ) can effectively be quantified by morphology
method.
Figure 2.8: Rough surface positioned on direct shear test
At 3S-R laboratory, a measurement system in contact surface known as asperities to a

certain degree has been experienced. Most of the ealier works were performed on the
measurement of the evolution of morphology of rock joints by using a laser beam (Armand et
al. [1998] , Hans & Boulon [2003]). This method is well appropriate to describe the asperities
surface. In accordance with this approach, the surface roughness can be directly measured.
This device as described in Figure 2.9 is composed of a laser scan fixed on a platform
motorized by two stepper motors generating two orthogonal displacements controlled by
two LVDTs from the PC. The morphology is obtained by scanning the surface by a series
of parallel profiles with a constant step. By this way a map of altitude of steel surface,
consisting in 128 profiles of each one 128 points for a maximum size of plate of 80 mm x 80
mm, is then obtained. Every scan of the surface is referred to the origin point of the plate
as shown in Figure 2.10. In this way the relative position of the origin point and each point
on the steel plate is precisely defined in the three orthogonal systems (X, Y, Z). Figure
2.11 shows typical measurement of the morphology of rough plate used in this study. The
surface roughness as defined by Rmax = 0.2 mm is the average measurement obtained from
morphology method. Taking into account the normalized roughness (Rn = Rmax /D50 )
proposed by Uesugi & Kishida [1986], with particle diameter D50 of Fontainebleau sand
(D50 = 0.23 mm), these rough plates then represent Rn = 0.87 which can be attributed to
be very rough. From available references on surface roughness definition, Rn ≥ 0.10 would
be able to stand for rough surface: Uesugi & Kishida [1986]; Uesugi et al. [1989]; Hu & Pu
[2004].
A comparison was also performed between the two different kinds of surface roughness. For
63
the steel plate (so-called ”smooth plate”), some of test series were additionally conducted.
All monotonic tests with smooth plate under CNL and CNS conditions describing the main
characteristics of interface will be prestented. The surface roughness of this plate delivered
from SOLCYP is intended to be similar to that of steel pile model tests. It is found
that Rmax of this plate is about 14 µm. This surface roughness has not been determined
by morphology method, but has been derived from SOLCYP. Considering the normalized
roughness (Rn ), the value of Rmax = 14 µm is very much less than D50 of Fontainebleau
sand. Therefore, it can be concluded that the surface of this plate (Rn = 0.06) is fairly
smooth (Figure 2.12).
Figure 2.9: View of the apparatus of morphology measurement: (1) Laser beam; (2) LVDT
displacement sensors of the laser beam (3) rough plate
Figure 2.10: Scheme of measurement of the morphology of surface roughness, Hans &
Boulon [2003]
64
Figure 2.11: Map of altitude of surface obtained from morphology of the rough mate-
rial,(left) shaded relief map, (right) 3D surface.
For the purpose of the sample preparation, having known the mean diameter of Fontainebleau
sand (D50 = 0.23 mm), the maximum gap between the shear box and bottom plate should
not be over 0.3 mm in order to prevent the leakage of fine particles of sand from this gap,
especially during the cyclic loading tests. Therefore, a spacing of 0.3 mm, created by a pair
of brass foils, was arranged between the plate and the 1/2 box (see Figure 2.13) during the
construction of the sand sample to the desired density. Furthermore, this gap was set in
order to avoid having the direct friction between half shear box and the rough plate.
Figure 2.12: Smooth plate delivered from SOLCYP
In this study all tests were performed as dry tests, i.e. completely drained tests. With much
less technical difficulty of direct shear device, the sample preparation began with the setup
of a pair of brass foils and the shear box, serving as a container of the material, on the rough
plate (see Figure 2.14). The sample was set up by simple pluviation in the air with a low-
high fall. Once the density index was fixed, the weight of sample was required to achieve the
65
desired sample height. The average height of the samples before application of σn was 20 ±
0.5 mm. At this stage of sample preparation, different fabrication processes of the sample
were used. Loose sample was achieved by a simple pluviation method in the air with a low-
high fall. The tamping and vibration methods for obtaining dense sample were therefore
used. After each layer of sand deposited by simple pluviation and having flattened the soil
surfacee, the load piston (top cap) was then placed. The sample underwent densification by
using the tamping and vibration methods. Before running the test, the load piston has not
to be in contact with the stirrup (load hanger) for applying the normal stress in order to
be sure that there is no load applying to the sample. After turning on the control system,
the stirrup was positioned on the load piston (top cap) which induced a small normal stress
acting on the sample (about 5-10 kPa). This small normal stress was also recorded. Then
the normal stress was continuously applied by the control system until reaching the required
value, after that shear stress was applied.
Figure 2.13: Setup of shear box with a pair of brass foils
a pair of brass foils
Figure 2.14: Preparation of a sample for interface shear test
66
2.4 Monotonic interface direct shear tests
2.4.1 Interpretation of interface shear test
As mentioned in the previous chapter, a shear band or the soil-structure interface cannot
explicitly be separated from the surrounding soil. Let us recall the direct shear test by using
stereophotogrammetry to display the evolution of the field of the soil-rough plate relative
displacement (Hoteit [1990]). Figure 2.15 displays the progressive localization of the field
of incremental relative displacements. At the beginning of shearing, a diffused convex zone
is created at the contract of the rough plate. Very quickly after this initial phase, this zone
turns into a precise thin layer of constant thickness within the shear box. This experiment
and some others by different authors motivate our subsequent choice by considering the
direct shear sample composed of two parts. During soil-structure direct shear test, the
sample is schematically composed of two parts subjected to two different stresses. Two
parts of direct shear test have to be interpreted by considering:
• The ”active” lower part being in contact with the rough material is a so-called inter-
face.
• The ”passive” upper part as the rest (most) of the sample overcoming this interface
is practically subjected to compression oedometer.
From the interface interpretation proposed by Boulon [1988], in case of constant σn (CNL),
the normal stress acting on the interface (σn′ ) and the normal relative displacement of inter-
face ([u’]) are considered within the generated interface (interface thickness, t ≈ 10D50 ) as
shown in Figure 2.16. The compression oedometer has no influence in this case. However,
on different paths (including CNS condition), it would be appropriate to consider an oe-
dometer correction of the normal relative displacement measured. Whereas σn is variable,
the variation of normal displacement of interface (∆[u′ ]) is due to the contribution of the
passive part (∆[u]oedometer ) and the variation of the sample height (∆h) as:
∆[u′ ] = ∆h − ∆[u]oedometer (2.3)
Boulon [1988] also performed an oedometer correction at CV condition and found that
there was no much difference on the maximum shear stress with the height of sample of 20
mm. For this reason, all components in this study can be considered as mean values within
the generated interface without taking into account the oedometer correction, according to
moderate changes in σn .
The normal stiffness of interface (kn int ) measured in the shear box is different from the
normal stiffness on the full sample of sand (kn sam ). This sfiffness is formed along the
interface and part of oedometer sample. The entire sample (sample height, h) is mobilized
during the implementation of the initial normal stress (path at constant normal stress, the
most common). However, only the interface (thickness t) is then mobilized when shearing
load is applied (Figure 2.17). The relationship between interface thickness (t) and sample
height (h) can be expressed by:
67
(a)
(b)
(c)
(d)
Figure 2.15: 2-D CNL test, Hostun sand-rough plate, σn = 100 kPa, Hoteit [1990]. Paral-
lelepipedic shear box (6 cm x 20 cm x 20 cm); (a) Shear stress - shear displacement curve (b)
Relative displacement vectors, increment 1-2 (c) Relative displacement vectors, increment
2-3 (d) Relative displacement vectors, increment 3-4
68
Figure 2.16: Direct shear test interpretation, Boulon [1988]
[u]int [u]sam h
= ⇒ kn int = kn sam (2.4)
t h t
where [u]int and [u]sam are the normal displacement of interface (t) and of sample (h),
respectively.
Or the normal stiffness (kn ) can be considered as;
h
∆σn = kn sam [u]sam = kn int [u]int = kn sam [u]int (2.5)
t
The normal stiffness (necessarily secant) will be taken into account in the calculation of
specific weight under applied stress. But the normal relative displacement can aso be taken
into account. Considering at beginning of the test, when the normal stress is applied, the
vertical displacement is then measured for the entire sample. The relative displacement of
interface is only proportional to t/h of the vertical displacement measured. At this stage,
the normal stiffness of interface (kn int ) resulting from an application of normal stress is
taken into account. The current unit weight, specific weight (γd0 σn ), of interface under an
applied normal stress is then expressed by
1
γd0 σn = γd0 σn =0 · (2.6)
σn
1− t·kn int
or it can be expressed in terms of proportion of normal displacement
1 1
γd0 σn = γd0 σn =0 · = γd0 σn =0 · (2.7)
[u]int σn [u]sam σn
1+ t 1+ h
where γd0 σn =0 is the initial unit weight without taking into account the normal stress.
[u]int σn and [u]sam σn are respectively the normal displacement of interface and of sample
due to an application of normal stress.
69
Vn = 0
Vn
[u]sam
oedometer part
sample (h) h-t
interface t [u]int
Figure 2.17: Interpretation of full sample and interface during the application of the initial
normal stress
[u]int_Vn = t*[u]sam_Vn / h time or N
[u]sam_Vn
[u]int_Vn – [u]sam_Vn
[u]int
[u] [u]sam
Figure 2.18: Full sample and interface during the application of the initial normal stress,
then during the monotonic or cyclic direct shear
However, when following whether nonotonic or cyclic direct shear tests (at constant normal
stress), the global normal displacement at any time is expressed as [u]sam , but only the in-
terface contributes to the normal relative displacement (Figure 2.18). During shear loading,
the additional relative displacement of interface coincides with the additional relative dis-
placement of the sample, but both relative normal displacements are different. Therefore,
the normal relative displacement of the interface at any given time or number of cycles,
[u]int , can be expressed by

t
[u]int = [u]sam − [u]sam σn · 1− (2.8)
h
In the following, for the sake of simplicity [u] stands for the normal relative displacement
of the interface instead of [u]int
70
2.4.2 Test results of monotonic interface shear test
Preliminarily to cyclic interface shear tests, monotonic interface shear tests on two different
values of roughness whether under CNL or CNS conditions, as shown in Table 2.3, were first
performed in order to determine the peak friction angle (δpeak ) between granular soil and
structure as well as the critical angle (δcrit ) and the characteristic angle (δcar ) separating the
contractive and dilative domains, and main characteristics of the interface. Two distinct
densities, dense (ID0 = 90 ± 3%) and loose sand (ID0 = 30 ± 3%) were performed under
CNL and CNS conditions. In case of CNS condition, three different values of the normal
stiffness (k = 1000, 2000 and 5000 kPa/mm) were performed.
Table 2.3: Tests for primary validation (monotonic tests)

ID0 type of plate γd0 Path σn
(%) (kN/m3 ) (kPa/mm) (kPa)
30 rough , smooth ≈ 14.74 0 (CNL) 60, 120, 310

90 rough , smooth ≈ 16.55 0 (CNL) 60, 120, 310
30 rough , smooth ≈ 14.74 1000(CNS) 60, 100, 310
30 rough ≈ 14.74 2000(CNS) 60, 100, 310
30 rough , smooth ≈ 14.74 5000(CNS) 60, 100, 310
90 rough , smooth ≈ 16.55 1000(CNS) 60, 100, 310
90 rough ≈ 16.55 2000(CNS) 60, 100, 310
90 rough , smooth ≈ 16.55 5000(CNS) 60, 100, 310
In accordance with the interface interpretation, the objective of accessing the density (spe-
cific gravity) of the soil-structure interface connected with the normal relative displacement
observed is necessary. However, there are different ways to access this approach. Some
attempts have been made to get a good correspondence between the interface thickness
and the density. Having known that the critical density depends only on the level of nor-
mal stress, each of these stress levels with the hypothesis of interface thikness which might
differ according to initial compaction should be considered. Since the normal relative dis-
placement is measured, the critical density could be expressed as a function of the normal
stress.
First considering rough plate, after the application of the normal stress, the vertical dis-
placement was then measured for the entire sample (Figure 2.19(a)). Indeed, the normal
displacement of interface is only in proportion to the interface thickness (t) and sample
height (h), which can be expressed as [u]int = [u]sam · (t/h). Considering the interface
thickness within the range of 10D50 − 12D50 , Figure 2.19(b) shows the normal displacement
of interface on both densities after applying the normal stress. These values of interface
thickness also led to the evaluation of critical density. Taking into account the interface
71
thickness (t = 10D50 ) on both densities, there was a small dispersion of the trend in critical
density as a function of initial normal stress. In the range of t = 10D50 − 12D50 , this small
dispersion still existed (Figure 2.20(a)). The value of 11D50 would be reasonable to repre-
sent the interface thickness. However, the interface thickness was then considered according
to Hammad [1991]’s investigation. This investigation showed that the thickness of the shear
band had a tendency to decrease when the initial density increased. This can be concluded
that the relative displacements as well as the rotations of grains in the loose sand-structure
interface are more independent than that in the case of the dense sand-structure interface.
Dejong et al. [2003] also reported that the thickness of the localized shear zone varied with
the test conditions and increased with the accumulative displacement. From monotonic
CNL tests, the interface thickness on dense and loose samples were then defined as 10D50
and 12D50 , respectively (Figure 2.20(b)) .
The interface test results on smooth suface are also presented in a similar way. Nevertheless,
the interpretation of interface thickness seems very difficult because smooth plate provides
very thin thickness or no interface thickness. In general, as can be seen in available observa-
tions on smooth plate, the slippage occurs along the contact surface between the structural
material and granular soil instead of the intense shear zone as can be seen on rough plate
(Uesugi & Kishida [1986]). Considering the Rn = 0.07 of this plate, even though this value
can be attributed to the smooth surface according to the criterion of surface classification
(Rcrit ≈ 0.1), this would provide fairly thin thickness. The interface thickness can then be
defined, in accordance with the critical density (Figure 2.21(b)), as 3D50 for dense sand and
9D50 for loose sand. It would be said that there is a good correspondence between smooth
and rough plate in critical density. Figure 2.21(a) shows the normal displacement of sample
[u]sam and of interface [u]int after applying initial normal stress. With very thin thickness of
interface on dense sand (3D50 ), the normal displacement due to the application of normal
stress is then very small.
0 0
-0.05
-0.01
-0.1
[u]sam (mm)
[u]int (mm)
-0.15 -0.02
monotonic CNL ; rough plate
I | 30
10int_30
D0 % ; t = 10D50
-0.2 ID0 | 30 % ; t = 11D50
11int_30
monotonic CNL ; rough plate -0.03 ID0 | 30 % ; t = 12D50
12int_30
-0.25 ID0 | 30 %
sample_ID30 ID0 | 90 % ; t = 10D50
10int_90
ID0 | 90 %
sample_ID90 ID0 | 90 % ; t = 11D50
11int_90
ID0 | 90 % ; t = 12D50
12int_90
-0.3 -0.04
0 50 100 150 200 250 300 350 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 2.19: Normal relative displacement measured due to application of normal stress:(a)
sample ; (b) interface with t = 10-12D50
72
16.4 16.4
monotonic CNL ; rough plate monotonic CNL ; rough plate
16.2 16.2
y (12D50) = 0.0021x + 15.375
R² = 0.9467
16.0 16.0 J d crit = 0.0023Vn + 15.285
Jd crit (kN/m3)
y (11D50) = 0.0023x + 15.346
Jd crit (kN/m3)
R² = 0.996 R² = 0.996
15.8 15.8
y (10D50) = 0.0025x + 15.311
R² = 0.9248
15.6 15.6
15.4 = 10D50
t10d50 15.4 D0 y 30% ; with t = 12D50
I12D50
= 11D50
t11d50 ,10D50
D0 y 90% ; with t = 10D50
= 12D50
t12d50
15.2 15.2
15.0 15.0
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 2.20: Critical density of interface shear tests according to monotonic CNL direct
shear tests on rough plate: (a) with different values of interface thickness ; (b) with t =
10D50 for dense sample and t = 12D50 for loose sample
0 16.4
-0.05 16.2
y = 0.0019x + 15.394
R² = 0.9902
16.0
-0.1
[u] (mm)
Jd (kN/m3)
15.8
-0.15
15.6
-0.2 monotonic CNL ;
monotonic CNL ; smooth plate smooth plate
15.4
ID0 | 30 % ; [u]sam
samp_30 ID0 | 30 % ; (t = 9D50)
samp_30
ID0 | 90 % ; [u]sam
samp_90 ID0 | 90 % ; (t = 3D50)
-0.25 samp_90
ID0 | 30 % ; [u]int (t = 9D50)
int_30 15.2
ID0 | 90 % ; [u]int (t = 3D50)
int_90
-0.3 15.0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Vn (kPa) Vn (kPa)
(a) (b)
Figure 2.21: Interpretation of monotonic CNL interface test on smooth plate: (a) normal
relative displacement measured due to application of normal stress ; (b) Critical density in
accordance with the interface thickness
Figure 2.22 shows the typical monotonic interface results under CNL condition with rough
plate on both densities. As can be seen, the interface shear tests on dense sample (ID0 ≈
90%) obviously showed dilative behaviour ([u] > 0) which was higher for σn = 60 kPa than
the other ones. It was also found that the shear stress increased with the shear relative
displacement [w] until reaching a peak value and then decreased (softening behaviour) to
a residual shear stress (critical state). With high σn , the softening phase was significantly
observed. The peak stress ratio (ηpeak = τpeak /σn ) for σn = 60 kPa was higher than that for
σn = 310 kPa. This is due to the decrease in dilation rate as a function of normal stress. In
case of loose sample (ID0 ≈ 30%), the shear stress increased continuously with relative shear
73
displacement [w] and had slight softening. The interface behaved contractively throughout
the test on loose sample.
300 300
ID0 | 90 % ; CNL ; ID0 | 30 % ; CNL ;
rough plate rough plate
250 V n = 60 kPa
ID90_sn60
250 VID30_sn60
n = 60 kPa
V n = 120 kPa
ID90_sn120 VID30_sn120
n = 120 kPa
200 V n = 310 kPa

ID90_sn310
200 VID30_sn310
n = 310 kPa
W (kPa)
W (kPa)
150 150
100 100
50 50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.2 0
0.15
-0.05
0.1
-0.1
>u@ (mm)
>u@ (mm)
0.05
ID0 | 90 % ; CNL ;
-0.15
rough plate ID0 | 30 % ; CNL ;
0
Vn = 60 kPa
ID90_sn60 rough plate
Vn = 120 kPa
ID90_sn120 Vn = 60 kPa
-0.2 ID30_sn60
-0.05 Vn = 310 kPa
ID90_sn310 Vn = 120 kPa
ID30_sn120
Vn = 310 kPa
ID30_sn310
-0.1 -0.25
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
1 1
IID0 || 90
90 %% ;; CNL
CNL ; ID0 | 30 % ; CNL ;
D
Vn = 60plate
rough kPa rough plate
0.8 VVn ==60
ID90_sn60
120kPa
kPa 0.8 Vn = 60 kPa
ID30_sn60
n
VVn ==120
310kPa
ID90_sn120 kPa Vn = 120 kPa
ID30_sn120
n
V = 310
ID90_sn310
n kPa Vn = 310 kPa
ID30_sn310
0.6
K WVn
0.6
K WVn
0.4 0.4
0.2 0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.22: Monotonic CNL interface tests with rough plate, σn = 60, 120 and 310 kPa
on dense sand (ID0 ≈ 90%) and on loose sand (ID0 ≈ 30%)
74
160 120
ID0 | 30% ; Vn = 120 kPa ; CNL test ID0 | 30% ; Vn = 120 kPa ; CNL test
140
end of shearing starting shearing 80
120
40
100
unloading slope ; ks
Vn (kPa)
W(kPa)
80 unloading slope ; kn 0
60
-40
40
applying Vn -80
20
0 -120
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0 1 2 3 4 5 6
[u] (mm) [w] (mm)
(a) (b)
Figure 2.23: Typical determination of interface stiffnesses on loose sample with rough plate,
σn = 120 kPa: (a) normal stiffness (kn ); (b) shear stiffness (ks )
On evaluating the stiffness of the interface, Figure 2.23 shows typical determination of
normal and shear stiffnesses on loose sample. The slopes of the unloading curves from σn -
[u] and τ -[w] diagrams give the values of kn and ks , respectively. However, as mentioned
above, the unloading slope of kn from σn -[u] diagram which is determined from the entire
sample has to be interpreted due to the normal displacement and the thickness of the
interface. While the interface thickness has no influence on shear stiffness, this stiffness can
directly be obtained from unloading curve of τ -[w] diagram.
Figure 2.24 also shows the typical monotonic CNL interface results with smooth plate
on both densities. Obviously, the surface roughness strongly influenced the interface test
results. The shear stress whether on dense sample or on loose sample increased with the
relative shear displacement [w] until reaching a peak value and then continued to a residual
shear stress (critical state) without showing any softening phase. As can be seen, the
interface shear tests show that the peak and critical values of shear stress with smooth
plate are obviously less than those with rough plate. It is found that the peak shear stress
observed from smooth plate is about 60 - 70 % of that from rough plate. This is due to
the effect of dilative behaviour found on rough plate. Smooth plate, contrary to rough
plate, did not show a significant dilation phase during shear loading. The normal relative
displacement [u] of smooth plate was then less contractive than that of rough plate.
In general, the interface behaviour under constant normal stress condition (CNL) is well
capable of describing the problems involving the slope stability and retaining walls. It
is very important to understand the condition in which the normal stress acting on the
interface varies during shear loading which can be found in case of pile foudations. In this
case, the soil adjacent to the pile induces a normal stiffnes (k), depending on the diameter of
pile (D) and shear modulus of soil (G) expressed as k = 4G/D, which leads to the variation
of normal stress acting on the interface. This variation of normal stress is attributed to
the tendency of soil whether to contract or to dilate. This shows the great interest of the
loading paths with an imposed stiffness.
75
250 200
ID0 | 90 % ; CNL ; ID0 | 30 % ; CNL ;
smooth plate smooth plate
Vn = 60 kPa
sn60
Vn = 60 kPa
sn60
200 Vn = 120 kPa
Vn = 120 kPa
sn120 150 sn120
Vn = 310 kPa
sn310 Vn = 310 kPa
sn310
150
W (kPa)
W (kPa)
100
100
50
50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.15 0
ID0 | 90 % ; CNL ;
smooth plate
V n = 60 kPa
sn60
0.1 V n = 120 kPa
sn120 -0.05
V n = 310 kPa
sn310
[u] (mm)
[u] (mm)
0.05 -0.1
ID0 | 30 % ; CNL ;
smooth plate
0 -0.15
Vn = 60 kPa
sn60
Vn = 120 kPa
sn120
Vn = 310 kPa
sn310
-0.05 -0.2
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.8 0.8
ID0 | 30 % ; CNL ;
0.7 0.7 smooth plate
Vn = 60 kPa
sn60
0.6 0.6 Vn = 120 kPa
sn120
Vn = 310 kPa
sn310
0.5 0.5
K WVn
K WVn
0.4 0.4
0.3 0.3
ID0 | 90 % ; CNL ;
0.2 smooth plate 0.2
V n = 60 kPa
sn60
0.1 Vn = 120 kPa
sn120 0.1
Vn = 310 kPa
sn310
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.24: Monotonic CNL interface tests with smooth plate, σn = 60, 120 and 310 kPa
on dense sand (ID0 ≈ 90%) and on loose sand (ID0 ≈ 30%)
76
Therefore, to illustrate how the interface behaves under CNS condition, the monotonic
interface shear tests under CNS condition were performed with the three different values
of normal stiffness (k = 1000, 2000 and 5000 kPa/mm) and initial normal stress σn0 = 60,
100 and 310 kPa. The effect of normal stiffness was also observed on two different values
of surface roughness.
Figure 2.25 illustrates the test results on dense sample (ID0 ≈ 90%) as well as on loose
sample (ID0 ≈ 30%) with initial normal stress (σn0 = 100 kPa), it was found that the
normal stiffness significantly influenced the interface behaviour. On dense sample, at the
beginning of shearing phase, the normal stress (σn ) slightly decreased due to contractive
phase while the shear stress (τ ) still evolved and afterward the normal stress associated
with shear stress increased as a result of dilative phase. The increase of stress state evolved
with the shear displacement [w] until reaching the peak value of stress ratio (τ /σn ) and
then decreased to the critical state. The increase of normal stiffness (k) caused a significant
increase in peak normal and shear stresses. Dense samples showed dilative behaviour which
decreased with increasing normal stiffness (k).
The behaviour of interface under CNS condition on loose samples was significantly different
from that observed on dense samples. When loose sand was subjected to shear loading
under CNS conditon, the main charateristic of the inteface was that the normal stress (σn )
associated with shear stress (τ ) decreased significantly. The effect of increasing the normal
stiffness (in this study k = 5000 kPa/mm) was evident. At the beginning of shearing
phase, the interface showed significant contraction and consequently led to the significant
degradation of normal stress. During shearing phase, the interface behaved contractively
and the degradation of normal stress accompanied with the degradation of shear stress
evolved continuously.
Comparisons between the tests performed under various conditions of the inital normal
stress (σn0 ) and normal stiffness (k) are also presented. Figure 2.26 and 2.27 show the
influence of initial normal stress (σn0 ) as well as the normal stiffness (k) on dense sand
(ID0 ≈ 90%) and on loose sand (ID0 ≈ 30%), respectively. On dense sand, the normal stress
associated with shear stress increased considerably as a function of normal stiffness. The
increase in normal stress (σn ) during shearing phase was due to the the dilative behaviour.
With imposing the low value of stiffness (k) the dilative behaviour was more evident. The
high value of imposed normal stiffness induces the reduction of dilation rate according to
the relation of k = ∆σ/∆[u]. On the other hand, loose sand significantly showed the
contractive behaviour throughout shearing phase which led to the significant degradation
of normal stress associated with shear stress. In particular, with σn0 = 60 kPa and k = 5000
kPa/mm, the stress state could only evolve in a small range ([w] ≈ 0.5 - 0.8 mm) due to the
rapid degradation of normal stress. With high value of imposed normal stiffness, in this case
k = 5000 kPa/mm which could be high enough to represent the constant volume condition
(CV), the normal relative displacement [u] of interface was almost constant throughout
shearing phase.
77
400 100
ID0 = 90% ; Vn0 = 100 kPa ID0 = 30% ; Vn0 = 100 kPa
kk ==1000
1000kPa/mm
kPa/mm kk == 1000
1000kPa/mm
kPa/mm
kk ==2000
2000kPa/mm
kPa/mm 80 kk == 2000
2000kPa/mm
kPa/mm
300 kk ==5000
5000kPa/mm
kPa/mm kk == 5000
5000kPa/mm
kPa/mm
60
W (kPa)
W (kP)
200
40
100
20
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
600 140
ID0 = 90% ; Vn0 = 100 kPa ; ID0 = 30% ; Vn0 = 100 kPa
kk == 1000
1000kPa/mm
kPa/mm k = 1000
1000kPa/mm
kPa/mm
500 k == 2000
2000kPa/mm
kPa/mm 120
k = 2000
2000kPa/mm
kPa/mm
k == 5000
5000kPa/mm
kPa/mm kk =
= 5000
5000kPa/mm
kPa/mm
100
400
Vn (kPa)
Vn (kPa)
80
300
60
200
40
100 20
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.2 0
ID0 = 90% ; Vn0 = 100 kPa ID0 = 30% ; Vn0 = 100 kPa
= 1000
kk = 1000kPa/mm
kPa/mm kk ==1000
1000kPa/mm
kPa/mm
0.15 = 2000
kk = 2000kPa/mm
kPa/mm kk ==2000
2000kPa/mm
kPa/mm
-0.02
k= = 5000
5000kPa/mm
kPa/mm kk ==5000
5000kPa/mm
kPa/mm
0.1
[u] (mm)
-0.04
[u] (mm)
0.05
-0.06
0
-0.08
-0.05
-0.1 -0.1
0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 8
[w] (mm) [w] (mm)
Figure 2.25: Monotonic CNS interface tests with three different normal stiffnesses (k), for
the ”rough plate”, σn0 = 100 kPa on dense sand (ID0 ≈ 90%) and on loose sand (ID0 ≈ 30%)
78
600 600
ID0 | 90 % ; CNS ; rough plate ID0 | 90 % ; CNS ; rough plate
k = 1000 kPa/mm k = 5000 kPa/mm
500 V
snn == 60
60kPa
kPa 500 Vn==60
sn 60kPa
kPa
V
snn == 100
100kPa
kPa Vn==100
sn 100kPa
kPa
V
snn == 300
310kPa
kPa Vn==300
sn 310kPa
kPa
400 400
W (kPa)
W (kPa)
300 300
200 200
100 100
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
700 700
ID0 | 90 % ; CNS ; rough plate
k = 1000 kPa/mm
600 Vn==6060kPa
kPa 600
sn
Vn==100
sn 100kPa
kPa
500 V = 310 kPa
snn= 300 kPa 500
Vn (kPa)
400 400
Vn (kPa)
300 300
200 200 ID0 | 90 % ; CNS ; rough plate

k = 5000 kPa/mm
V
snn == 60
60kPa
kPa
100 100 V
snn == 100
100kPa
kPa
V
snn == 300
310kPa
kPa
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.2 0.15
0.15
0.1
0.1
[u] (mm)
0.05
[u] (mm)
0.05
0
0 ID0 | 90 % ; CNS ; rough plate
ID0 | 90 % ; CNS ; rough plate k = 5000 kPa/mm
k = 1000 kPa/mm V
snn==60
60kPa
kPa
Vn==6060 kPa -0.05 V
-0.05 sn kPa snn==100
100kPakPa
Vn==100
sn 100
kPakPa V
snn==300
310kPakPa
Vn==300
sn 310
kPakPa
-0.1 -0.1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.26: Monotonic CNS interface tests with rough plate on dense sand (ID0 ≈ 90%),
k = 1000 and 5000 kPa/mm
79
250 250
k = 1000 kPa/mm k = 5000 kPa/mm
Vn==60
60kPa
kPa Vn ==60
sn 60kPa
kPa
200 sn 200
Vn==100
sn 100kPa
kPa Vn ==100
sn 100kPa
kPa
Vn==300
sn 310kPa
kPa Vn ==300
sn 310kPa
kPa
150 150
W (kPa)
W (kPa)
100 100
50 50
0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
400 400
k = 1000 kPa/mm k = 5000 kPa/mm
Vn==60
sn 60kPa
kPa V
snn ==60
60kPa
kPa
300 Vn==100
sn 100kPa
kPa 300 V
snn ==100
100kPa
kPa
Vn==300
sn 310kPa
kPa V
snn ==300
310kPa
kPa
Vn (kPa)
Vn (kPa)
200 200
100 100
0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.05 0.05
k = 1000 kPa/mm
V
snn == 60
60kPa
kPa
V 0
0 snn == 100
100kPa
kPa
V
snn == 300
310kPa
kPa
[u] (mm)
-0.05
[u] (mm)
-0.05
-0.1
-0.1 k = 5000 kPa/mm
-0.15 V
snn ==60
60kPa
kPa
V
snn ==100
100kPa
kPa
V
snn ==300
310kPa
kPa
-0.15 -0.2
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.27: Monotonic CNS interface tests with rough plate on loose sand (ID0 ≈ 30%), k
= 1000 and 5000 kPa/mm
80
400 120
ID0 = 90% ; Vn0 = 100 kPa ID0 = 30% ; Vn0 = 100 kPa
kk = 1000
1000kPa/mm
kPa/mm 1000kPa/mm
k = 1000 kPa/mm
kk = 2000 kPa/mm
2000 kPa/mm 100 2000kPa/mm
k = 2000 kPa/mm
kk = 5000 kPa/mm
5000 kPa/mm 5000kPa/mm
k = 5000 kPa/mm
300
80
Kcrit = 0.566
W (kPa)
W (kPa)
200 60 1
Kcrit = 0.566
1
40
100
20
0 0
0 100 200 300 400 500 0 20 40 60 80 100 120 140 160
Vn (kPa) Vn (kPa)
(a) (b)
Figure 2.28: Stress paths of monotonic CNS interface tests with rough plate, σn0 = 100
kPa; (a) dense sand (ID0 ≈ 90%); (b) loose sand (ID0 ≈ 30%)
400 600
k = 1000 kPa/mm k = 5000 kPa/mm
V
snn==60
60kPa
kPa 500 Vn==6060kPa
sn kPa
Vn==100
sn 100kPa
kPa Vn==100
sn 100 kPa
kPa
300
Vn==300
sn 310kPa
kPa Vn==300
sn 310 kPa
kPa
400
W (kPa)
W (kPa)
200 300
Kcrit = 0.566
Kcrit = 0.566
1
200 1
100
100
0 0
0 100 200 300 400 500 0 100 200 300 400 500 600 700
Vn (kPa) Vn (kPa)
300 300
k = 1000 kPa/mm k = 5000 kPa/mm
Vn==6060kPakPa
sn Vn==60
sn 60kPa
kPa
Vn==100
sn 100kPakPa Vn==100
sn 100kPa
kPa
Vn==300
sn 310kPakPa Vn==300
sn 310kPa
kPa
200 200
W (kPa)
Kcrit = 0.566 Kcrit = 0.566

W (kPa)
1 1
100 100
0 0
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
Figure 2.29: Stress paths of monotonic CNS interface tests with rough plate, k = 1000 and
5000 kPa/mm
81
Figure 2.28 and 2.29 also show the stress paths of monotonic CNS tests with various initial
conditions. Even though different values of normal stiffness (k) were applied, the interface
tests still mobilized the same critical stress ratio (ηcrit = τcrit /σn ≈ 0.566). This observation
is in accordance with that observed by Tabucanon et al. [1995]. The peak shear stress ratio
(ηpeak = τpeak /σn ) slighly decreased with an increase function of normal stress and an
imposed normal stiffness. The stress paths of different values of normal stffness whether on
dense sand or loose sand also showed the similar trend of stress ratio (η) after reaching the
peak stress ratio ηpeak which lied on the critical stress ratio (ηcrit ).
To verify the regularity of imposed normal stiffness, Figure 2.30 shows the relationship
between the variation of normal stress ∆σn and of normal relative displacement ∆[u] during
shear loading. The scatter of these tests resulted from the applied normal stress from
generating engine in two directions as described in the section of interface direct shear
devices. It should be noticed that the loose tests (CNL as well as CNS) are less stable,
due to the occurrence of oscillation, than dense tests. This was evident when performing
the tests with low inital normal stress (σn0 = 60 kPa) and high value of imposed normal
stiffness (k = 5000 kPa/mm).
Figure 2.31 shows the critical and peak stress envelopes of the test results with rough plate
on both densities. The envelopes of these stress states for CNS tests corresponded well to
that for CNL tests. These envelopes were fit with the power functions as can also be seen
in Mortara [2001]. It would be said that the value of the normal stiffness has no influence
on these envelopes.
400 800
Vn0 = 60 kPa ; CNS tests ; Vn0 = 310 kPa ; CNS tests ;
350 rough plate 700 rough plate
|30= 1000
IIDD0= 30%_k % ;kPa/mm
k = 1000 kPa/mm kID0 |30kPa/mm
= 1000 %; k = 1000 kPa/mm
300 |
IIDD0= 30%_k = 5000;kPa/mm
k = 5000 kPa/mm 600 kID0 |
= 5000 ; k = 5000 kPa/mm
kPa/mm
D0 |0
IID=90%_k % kPa/mm
= 1000 ; k = 1000 kPa/mm ID0 |0 % ; k = 1000 kPa/mm
ID90_1000k
250 D0 |%
IID=90%_k ; k = 5000 kPa/mm
= 5000 kPa/mm
500 ID0 |% ; k = 5000 kPa/mm
ID90_5000k
Vn (kPa)
Vn (kPa)
k = UVn/U[u] = 5000 kPa/mm

200 400
150 300
100 200
50 100
0 0
-0.16 -0.12 -0.08 -0.04 0 0.04 0.08 0.12 0.16 -0.15 -0.1 -0.05 0 0.05 0.1 0.15
[u] (mm) [u] (mm)
(a) (b)
Figure 2.30: Relationship of imposed normal stiffness during monotonic tests with rough
plate on loose and dense sands: (a) σn0 = 60 kPa ; (b) σn0 = 310 kPa
82
300 300
ID0 | 90 % ; CNL and CNS ; rough plate ID0 | 30 % ; CNL and CNS ; rough plate
y = 0.8254x0.9482
R² = 0.998
200 200
y = 0.9512x0.9739 »peak
o°condition
¤¼¨condition
W (kPa)
W (kPa)

R² = 0.9994 »critical
o°¤¼(power)
¨ y = 0.776x0.943
y = 0.7871x0.9432 ¥Î
fittingµ¨´»o°¤¼¨ R² = 0.9941
R² = 0.9967 ¥Î
fitting »o°¤¼¨
µ¨´(power)
100 100
»peak
o°¤¼condition
¨
»critical
o°¤¼¨condition

¥Î µ ¨´
»o°¤¼¨
fitting (power)
¥Î µ¨´»o°¤¼¨
fitting (power)
0 0
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 2.31: Envelopes of peak and critical stress states for monotonic CNL and CNS
interface shear tests with rough plate : (a) dense sand (ID0 ≈ 90%) ; (b) loose sand
(ID0 ≈ 30%)
With smooth plate, the test results also showed the influence of imposed normal stiffness. A
similar trend, as can be observed with rough plate, was expected during shearing. However,
it was found that the tests results obtained from smooth plate were different from those
obtained from rough plate. As can generally be observed from monotonic CNL tests on
smooth plate as well as rough plate, the surface roughness strongly influenced the test
results. Especially under CNS condition, the dilative behaviour which could significantly
be observed, depending on σn0 as well as k, on rough plate provided the significant variation
of normal stress associated with shear stress. On the other hand, Figure 2.32 shows that
smooth plate did not show the significant dilation and therefore the stress state did not
show significant variation. The normal stress slighly increased during shearing phase (this
could only be found in case of σn0 = 310 kPa and k = 5000 kPa/mm). Consequently, the
shear stress evolved continuously whithout showing softening phase. Figure 2.33(a) also
shows stress paths of monotonic CNS tests on dense sand. The different values of imposed
normal stiffness (k = 1000 and 5000 kPa/mm) slighly influenced the interface behaviour
on smooth plate. With moderate value of initial normal stress (σn0 = 60 and 100 kPa)
the different values of k seemed to have no influence on the test results. It was clear that
the volumetric behaviour of interface was not influenced by the imposed normal stiffness.
The normal relative displacement ([u]) from various conditions were almost constant during
shearing phase. The dilative behaviour could hardly be observed on smooth plate.
On loose sand, it was found that the influence of imposed normal stiffness observed on
smooth plate was similar to rough plate (the significant degradation in σn ). During shearing,
the interface behaved contractively and then the degradation of normal stress increased
significantly. This also led to the degradation of shear stress. As can be seen on rough plate,
the rate of normal stress degradation significantly increased as a function of imposed normal
stiffness on loose sand. It was also found that the rate of stress degradation on smooth plate
was higher than rough plate. Obviously, on smooth plate with k = 5000 kPa/mm, σn0 =
83
310 kPa the normal stress decreased from 310 kPa to 100 kPa in the range of [w] = 1 mm
whereas on rough plate the normal stress decreased from 310 kPa to 200 kPa. The stress
paths on smooth plate (Figure 2.33) also showed the trend of stress degradation rate when
increasing the imposed normal stiffness. The normal relative displacement under various
conditions on smooth plate showed that the imposed normal stiffness had no influence on
the volumetric behaviour of the interface.
400 200
ID0 | 90% ; CNS ; smooth plate ID0 | 30% ; CNS ; smooth plate
V n0 = 60
sn60_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm V n0 = 60
sn60_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm
V n0 = 60
sn60_k kPakPa/mm
= 5000 ; k = 5000 kPa/mm V n0 = 60
sn60_k kPakPa/mm
= 5000 ; k = 5000 kPa/mm
300 V n0 = 100
sn100_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm 150 V n0 = 100
sn100_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm
V n0 = 100
sn100_k kPakPa/mm
= 5000 ; k = 5000 kPa/mm V n0 = 100
sn100_k kPakPa
= 5000 ; k = 5000 kPa/mm
V =
sn310_k
n0 310 kPa ; k = 1000 kPa/mm
= 1000 kPa/mm V =
sn310_k
n0 310 kPa ; k = 1000 kPa/mm
= 1000 kPa/mm
V n0 = 310
sn310_k kPakPa/mm
= 5000 ; k = 5000 kPa/mm V =
sn310_k
n0 310
= kPa
5000 ; k = 5000 kPa/mm
kPa/mm
W (kPa)
W (kPa)
200 100
100 50
0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
[w] (mm) [w] (mm)
500 400
ID0 | 30% ; CNS ; smooth plate
Vn0 = 60
sn60_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm
Vn0 = 60
sn60_k kPakPa/mm
= 5000 ; k = 5000 kPa/mm
400 Vn0 = 100 kPakPa/mm
; k = 1000 kPa/mm
300 sn100_k = 1000
Vn0 = 100
sn100_k kPakPa
= 5000 ; k = 5000 kPa/mm
Vn0 = 310
sn310_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm
300 Vn0 = 310
sn310_k kPakPa/mm
= 5000 ; k = 5000 kPa/mm
Vn (kPa)
Vn (kPa)
200
200
100
100
0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
[w] (mm)
[w] (mm)
0.02 0.1
0.00
0
-0.02
[u] (mm)
[u] (mm)
-0.1
-0.04
ID0 | 90% ; CNS ; smooth plate
-0.2 ID0 | 30% ; CNS ; smooth plate
V n0 = 60
sn60_k = kPa ; k = 1000 kPa/mm
1000 kPa/mm
V n0 = 60
sn60_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm
-0.06 V n0 = 60
sn60_k = kPa ; k = 5000 kPa/mm
5000 kPa/mm
V n0 = 60
sn60_k kPakPa/mm
= 5000 ; k = 5000 kPa/mm
V =
sn100_k100= kPa
1000; k = 1000 kPa/mm
kPa/mm
n0 V =
sn100_k100
= kPa
1000 ; k = 1000 kPa/mm
kPa/mm
V =
sn100_k100= kPa
5000; k = 5000 kPa/mm
kPa/mm -0.3
n0
-0.08
n0 V n0 = 100
sn100_k kPakPa
= 5000 ; k = 5000 kPa/mm
V n0 = 310
sn310_k kPa ;kPa/mm
= 1000 k = 1000 kPa/mm
V n0 = 310
sn310_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm
V n0 = 310
sn310_k kPa ;kPa/mm
= 5000 k = 5000 kPa/mm
V n0 = 310
sn310_k kPakPa/mm
= 5000 ; k = 5000 kPa/mm
-0.10 -0.4
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
[w] (mm) [w] (mm)
Figure 2.32: Monotonic CNS interface tests with smooth plate, k = 1000 and 5000 kPa/mm
84
400 300
ID0 | 90% ; CNS ; smooth plate ID0 | 30% ; CNS ; smooth plate
V Vn0 = 60
sn60_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm
n0 == 60
sn60_k 1000kPa
kPa/mm ; k = 1000 kPa/mm
V
sn60_k== 60
5000kPa
kPa/mm ; k = 5000 kPa/mm Vn0 = 60
sn60_k kPakPa/mm
= 5000 ; k = 5000 kPa/mm
n0
300 V Vn0 = 100
sn100_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm
n0 = 100 kPa ; k = 1000 kPa/mm
sn100_k = 1000 kPa/mm
V Vn0 = 100
sn100_k kPakPa
= 5000 ;k = 5000 kPa/mm
n0 = 100 kPa ; k = 5000 kPa/mm
sn100_k = 5000 kPa/mm
200 Vn0 = 310
V n0 = 310 kPa ; k = 1000 kPa/mm
sn310_k = 1000 kPa/mm sn310_k kPakPa/mm
= 1000 ; k = 1000 kPa/mm
V Vn0 = 310 kPakPa/mm
; k = 5000 kPa/mm
W (kPa)
=
sn310_k =310 kPa ; k = 5000 kPa/mm
5000 kPa/mm sn310_k = 5000
W (kPa)
n0
200
Kcrit = 0.43
1 Kcrit = 0.43
100 1
100
0 0
0 100 200 300 400 500 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 2.33: Stress paths of monotonic CNS interface tests with smooth plate, k = 1000
and 5000 kPa/mm : (a) dense sand (ID0 ≈ 90%) ; (b) loose sand (ID0 ≈ 30%)
After finishing the test, additional observation of the grain breakage within interface shear
zone was carried out. Generally, shear loading causes a substantial disruption of grains
within the shear zone which depends on several factors such as grain size, grain shape,
σn0 , ID0 and the surface roughness of structure. From many observations relevant to soil-
structure interface and the measurement of interface thickness in this study, it is evident
that the approximate interface thickness for granular materials is about 10 − 12D50 . In
this study, with D50 = 0.23 mm of Fontainebleau sand resulting from sieving method, the
approximate layer of 2 - 3 mm, depending on the initial density and type of plate, of sand
adjacent to the plate was therefore carried out by means of sieving method.
The extension of this purpose is to describe the specific energy (W ) which allows a first
approach of the level of grain breakage within the interface during shear loading. It is
expressed by using the following formula;
Z
W = (τ · [ẇ] − σn [u̇]) dtime (2.9)
time
This energy is also used in the modeling of interface behaviour in the form of nonlinear
incremental law (Garnica-Anguas [1993]). It is comprehensible that with sufficient energy
induced during shear loading the particles within the localized shear zone can be crushed.
The grain breakage also depends on the crushability of grains (Uesugi et al. [1989]). Figure
2.34 and 2.35 show typical curves representing the specific energy from interface shear tests
on rough and smooth plates, respectively. For CNL tests on rough and smooth plates, the
specific energy increased as a function of σn and it was found that this specific energy was
nearly independent of ID0 . As can be seen, the smooth plate obviously provided less specific
energy than rough plate.
85
2000 2000
ID0 | 90 % ; CNL ; ID0 | 30 % ; CNL ;
rough plate rough plate
Vn = 60 kPa
ID90_sn60 Vn = 60 kPa
ID30_sn60
1500 Vn = 120 kPa
ID90_sn120 1500 Vn = 120 kPa
ID30_sn120
Vn = 310 kPa
ID90_sn310 Vn = 310 kPa
ID30_sn310
W (Joules/m2)
W (Joules/m2)
1000 1000
500 500
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
2000 250
ID0 = 90% ; Vn0 = 100 kPa
ID0 = 30% ; Vn0 = 100 kPa
= 1000kPa/mm
k =1000 kPa/mm
k = 1000kPa/mm
k=1000 kPa/mm
k = 2000
k=2000 kPa/mm
kPa/mm
200 k = 2000kPa/mm
k=2000 kPa/mm
k = 5000
k=5000 kPa/mm
kPa/mm
1500 k = 5000kPa/mm
k=5000 kPa/mm
W (Joules/m2)
W (Joules/m2)
150
1000
100
500
50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.34: Specific energy from monotonic CNL and CNS tests on rough plate
1600 1600
ID0 | 30 % ; CNL ; ID0 |I90
D | 30 % ; CNL
% ; CNL ;
1400 smooth plate 1400 Vn =plate
smooth 60 kPa
V
n = 60 kPa Vn = V
60n =kPa
120 kPa
V

n = 120 kPa Vn = V
n = kPa
120 310 kPa
1200 1200
V

n = 310 kPa Vn =
310 kPa
W (Joules/m2)
W (Joules/m2)
1000 1000
800 800
600 600
400 400
200 200
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
(a) (b)
Figure 2.35: Specific energy from monotonic CNL tests on smooth plate: (a) loose sand
(ID0 ≈ 30%) ; (b) dense sand (ID0 ≈ 90%)
86
For CNS tests, dense sand also showed an increase of specific energy as a function of normal
stiffness (k) due to the siginificant increase in normal stress associated shear stress. On
the other hand, loose sand showed different trend of specific energy. An increase of the
normal stiffness for the same initial normal stress induced the degradation of stress. The
shear stress passed a peak value and then decreased, while the normal stress significantly
decreased during shear loading. This can be said that with small amount of specific energy
the particles within the inteface would not be crushed (Uesugi et al. [1989]).
Figure 2.36 - 2.38 show typical grain size distribution curves after testing by means of
sieving method. On rough plate, it turned out that the particle size (D50 ) after testing
under CNL condition did not changed so much, in comparison with the curve of initial
state resulting from sieving method, (see Figure 2.36) even with high initial normal stress
and high initial density. However, when considering the fine particle size (D10 ), at high
value of normal stress (σn = 310 kPa), there were small variations of fine particle size
(D10 ), a shif of grading curves to the left. This variation from initial state on dense sample
was more evident than on loose sample. Even though the specific energy of dense sand was
almost the same as loose sand but the tendency of grain to be crushed on dense sand would
be higher than that on loose sand. With low initial normal stress (σn = 60 kPa) especially
on low density, the grain breakage could hardly be observed.
100 100
Initial (laser method)
laser
90 90 Initial (sieving method)
initial
finer by weight (%)
ID0 = 30% ; Vn = 310 kPa

Loos60_mono
80 80 ID0 = 90% ; Vn = 310 kPa
dense60_mono
finer by weight (%)
70 70
60 60
laser
50 50
Initial (sieving method)
initial
40 ID0 = 30% ; Vn = 60 kPa
Loos60_mono 40
ID0 = 90% ; Vn = 60 kPa
dense60_mono
30 30
20 20
10 10
0 0
0.01 0.1 1 10 0.01 0.1 1 10
particle size (mm) particle size (mm)
(a) (b)
Figure 2.36: Grain size distribution of sand after testing of monotonic CNL tests on rough
plate: (a) with normal stress σn = 60 kPa ; (b) with normal stress σn = 310 kPa
Under CNS condition, the variation of normal stress associated with the shear stress induced
the grain breakage within the localized shear zone differing from that as observed under
CNL condition. On dense sand, with increasing the value of imposed normal stiffness (k)
the dilative behaviour which could be observed with rough plate resulted in the significant
increase in normal stress associated with shear stress. Therefore, the specific energy as a
product of shear stress and shear displacement increased significantly. However, grain size
distribution curves still provided the small vriation of particle size (D50 ). Following the
same procedure in which the the fine particle (D10 ) was considered, the variation of D10
observed from CNS tests was more evident than that from CNL condition. The change in
87
fine particle size described by D10 was considerable for increasing the initial normal stress
and imposed normal stiffness as shown in Figure 2.37.
On the other hand, loose sand showed the contractive behaviour throughout shearing tests
which consequently led to the degradation of normal stress associated with shear stress.
Therefore, the particle within interface shear zone could not considerably be crushed. With
σn0 = 310 kPa, the variation of fine particle size (D10 ) could fairly be observed. For low
initial normal stress σn0 = 60 kPa, the interface tests which could only performed in a small
range ([w] ≈ 1 mm) due to the significant reduction of stress state showed no variations in
grain size distribution curves.
100 100
90 90
80 80
finer by weight (%)
finer by weight (%)

70 70
60 60
50 50
40 40
ID0 = 30% ; Vn = 60 kPa ID0 = 90% ; Vn = 60 kPa
30 30
initial initial
initial
20 monotonic CNL
loose60_mono 20 monotonic CNL
dense60_mono
1000k
monotonic k = 1000 kPa/mm monotonic k = 1000 kPa/mm
1000k
10 5000k
monotonic k = 5000 kPa/mm 10 monotonic k = 5000 kPa/mm
5000k
0 0
0.01 0.1 1 10 0.01 0.1 1 10
100 100
90 90
80 80
finer by weight (%)
70 70
finer by weight (%)
60 60
50 50
40 40
30 ID0 = 30% ; Vn = 310 kPa 30 ID0 = 90% ; Vn = 310 kPa
20 initial
initial 20 initial
monotonic CNL
loose310_mono monotonic CNL
dense310_mono
10 monotonic k = 1000 kPa/mm
1000k 10 1000k
monotonic k = 1000 kPa/mm
5000k 5000k
0 0
0.01 0.1 1 10 0.01 0.1 1 10
Figure 2.37: Grain size distribution of sand after testing of monotonic CNS tests on rough
plate with initial normal stress σn0 = 60 and 310 kPa
On smooth plate, the trend of grain size distibution curves was not much different from
that observed on rough plate. The fine particle size (D10 ) was still considered. It was
found that the trend of grain breakage on smooth plate was less evident than rough plate.
With σn0 = 60 kPa the change of fine particle size from the initial state (a shift of grading
curve) could hardly observed. This can be attributed to the interface mechanism on two
distinct surfaces. If the stress state is high enough, the localized shear zone constitued along
the sand-rough plate interface induces grain breakage (Uesugi et al. [1989]). In the same
88
condition, sand-smooth plate that the slippage normally occurrs along the contact surface
shows less grain breakage.
100 100
90 90
80 80
finer by weight (%)
70 70
finer by weight (%)

60 60
50 50
laser Initial (laser method)
laser
40 Initial (sieving method)
initial 40 Initial (sieving method)
initial
ID0 = 30% ; Vn = 60 kPa
Loos60_mono ID0 = 30% ; Vn = 310 kPa
loose_310
30 ID0 = 90% ; Vn = 60 kPa
dense60_mono 30 ID0 = 90% ; Vn = 310 kPa
dense310_mono
20 20
10 10
0 0
0.01 0.1 1 10 0.01 0.1 1 10
(a) (b)
Figure 2.38: Grain size distribution of sand after testing of monotonic CNL tests on smooth
plate: (a) with normal stress σn = 60 kPa ; (b) with normal stress σn = 310 kPa
From these monotonic tests under CNL condition, the peak friction (δpeak ), critical (δcrit )
and characteristic (δcar , separating dilative and contractive domain) angles were then de-
termined. These angles were used to design the cyclic test campaign. It was found that the
peak friction angle obviously depended on the initial density as well as the surface rough-
ness. Figure 2.39 and 2.40 respectively show the peak shear stresses (τpeak ) of all tests in
τ − σn plane of rough and smooth plates. The peak, critical and characteristic lines arose
from the fitting of a linear function through the origin. The peak friction angle (δpeak ) was
calculated from the ratio between the maximum shear stress along the inclusion and initial
normal stress (Schlosser & Guilloux [1981]) as:
τpeak
tan δpeak = (2.10)
σn
Then the peak friction angle with rough plate on dense and loose samples δpeak = 38.3◦ (ηpeak
= 0.79) and δpeak = 31.4◦ (ηpeak = 0.61) were approximately held, respectively. Similarly,
the critical (δcrit ) and the characteristic angles (δcar ) which were taken into account as the
average values for dense and loose samples were δcrit = 29.5◦ (ηcrit = 0.566) and δcar = 29◦
(ηcar = 0.555), respectively. These two angles seemed to be the same value. According to
the definition of the characteristic angles (δcar ) proposed by Luong [1980], it is found to be
independent of the density and the confined stress. The characteristic and critical states
were very similar. For loose sand at high confined stress the failure which generally occurs
at the critical state would be attributed to the characteristic state as well.
For smooth surface, the peak friction angle on dense and loose samples δpeak = 26.1◦ (ηpeak
= 0.49) and δpeak = 23.3◦ (ηpeak = 0.43) were held, respectively. In the same way, the
critical and characteristic angles on dense sameple were 23.3◦ (ηcrit = 0.43) and 18.8◦ (ηcar
89
= 0.34), respectively. Because smooth plate did not show softening behaviour, the peak and
critical values of stress ratio were almost the same on loose sample. Therefore, the critical
angle on loose sample was 23.3◦ (ηcrit = 0.43).
300 300
ID0 | 90 % ; CNL ; rough plate ID0 | 30 % ; CNL ; rough plate
Kpeak = 0.79
200 1 200
W (kPa)
W (kPa)
Kcrit = 0.566 Kpeak = 0.61
1
1
100 100
Kcar = 0.555 Kcrit = 0.566
1 1
0 0
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 2.39: Peak and characteristic stress in τ − σn plane from monotonic CNL interface
tests on rough plate; (a) dense sand (ID0 ≈ 90%); (b) loose sand (ID0 ≈ 30%)
300 300
ID0 | 90 % ; CNL ; smooth plate ID0 | 30 % ; CNL ; smooth plate
200 200
W (kPa)
W (kPa)
Kpeak = 0.49
1
100 100 Kpeak = 0.43

Kcrit = 0.43
1
1
Kcrit = 0.43
Kcar = 0.34 1
1
0 0
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 2.40: Peak and characteristic stress in τ − σn plane from monotonic CNL interface
tests on smooth plate; (a) dense sand (ID0 ≈ 90%) ; (b) loose sand (ID0 ≈ 30%)
It was also found that the peak interface friction angle (δpeak ) on rough plate was com-
paratively lower than the internal friction angle of Fontainebleau sand (ϕpeak ) as shown in
Figure 2.41. These internal friction angles of Fontainebleau sand were obtained from avail-
able observations in triaxial tests. Tabucanon et al. [1995] also reported that the interface
friction angle was lower than internal friction angle on standard shear box for Bass Strait
calcareous sand, however both friction angles mobilized the same critical friction angle.
They concluded that the shear resistance of interface was controlled by the deformation
90
of sand adjacent to the rough plate which was lower than that mearsured in sand-sand on
standard shear box. For smooth plate the interface friction angle was obviously lower than
those obtained from rough pate and the internal friction angle. This could be said that
on smooth plate the sand slipped at the interface and the volume change could hardly be
observed.
50
triaxial LCPC Nantes Gérard B1
triaxial DGI B1
45
interface direct shear (rough plate)
interface direct shear (smooth plate)
Mpeak Gpeak (degree)
40
35
30
25
20
14.5 15.0 15.5 16.0 16.5 17.0
Jd (kN/m3)
Figure 2.41: Peak interface friction angle (δpeak ) with rough and smooth plates in compar-
ison with peak internal friction angle (ϕpeak ) of Fontainebleau sand
2.5 Test parameters and extracted variables from the tests
The objective of this study is to clarify the main trends (extrapolated to other interfaces) of
the behaviour of granular soil-structure interface under cyclic loading with a large number of
cycles. In order to formulate the results of this test campaign in a visco-plastic constitutive
model in which the number of cycles (N ) is the fictitious time (t), and the mean cyclic
path during the regular cycles (N > 1) is considered more than the detail of each cycle,
see Figure 2.42. So far the most salient observations in cyclic loading domain with a large
number of cycles (e.g. Niemunis et al. [2005], Wichtmann [2005], Wichtmann et al. [2005])
show that the explicit method is suitable to describe the accumulation of granular soil
deformation. The fist (irregular) cycle and the subsequent (regular) cycles are different due
to the deformation at the beginning of cyclic loading. Therefore, the regular cycles (N ≥ 2)
are more representative for an analysis of the accumulated deformation.
All the tests are set at N = 104 cycles, at least if the failure is not reached before this
objective. Moreover, as indicated before, the cycles are carried out between two shear stress
thresholds (shear stress-controlled test). The data aimed in these tests are the mean cyclic
paths or irreversible relative displacements as well as the interface stiffness modification, and
finally the interface resistance modification during the cycles. In the τ − σn , the interface
variables used in the cyclic test campaign are therefore defined as (Figure 2.43):
91
Local cyclic path
Local mean cyclic path
t,N
(a) (b)
Figure 2.42: (a) mean cyclic path for cyclic interface shear test; (b) calculation procedure
of explicit models, Wichtmann et al. [2005]
• δpeak and ηpeak are respectively the peak friction angle (resistance) and the peak stress
ratio.
• δcar (0 < δcar < δpeak ) is the characteristic angle of soil-structure interface (separating
cyclic contracting and dilating domains).
• σn cm is the level of mean cyclic normal stress in CNL tests and σn cm0 is the level of
initial mean cyclic normal stress in CNS tests.
• τcm is the level of mean cyclic shear stress.
• ∆τ is the amplitude of the cycles in terms of shear stress.
• ηcm is the level of mean cyclic stress ratio.
In the following, the subscript ”cm” denotes the mean cyclic path. The soil-structure friction
angle (δ) and mean cyclic stress ratio (ηcm ) are expressed as:
τcm
ηcm = tan δcm = (2.11)
σn cm
The possible maximum stress ratio (ηmax ) that can be carried out in the campaign of
experimental observation is defined as:
∆η
ηcm + 2
ηmax = ≤ ηpeak (2.12)
1− ηcm ∆η
2
∆τ
where ∆η = σn cm
92
Gcrit Gcar
Kcar
Kcrit
-Kcar
-Kcrit
-Gcrit -Gcar
Figure 2.43: Characterization of the implemented cycles
The expected results were the displacement vectors as mean cyclic tangential [w]cm and
mean cyclic normal [u]cm relative displacements. In addition, the evolution of normal kn
and tangential ks stiffnesses as well as interface resistance δpeak of the interface after cyclic
shear loading were also expected:
[w]cm = [w]cm (ID0 , σn cm , ηn cm , N ) (2.13)

[u]cm = [u]cm (ID0 , σn cm , ηn cm , N ) (2.14)
kn = kn (ID0 , σn cm , ηn cm , N ) (2.15)
ks = ks (ID0 , σn cm , ηn cm , N ) (2.16)
δpeak = δpeak (ID0 , σn cm , ηn cm , N ) (2.17)
Finally, the grain breakage within interface shear zone was then observed by means of
sieving method.
2.6 Operating mode and experimental biases
The procedure of cyclic test campaign in this study follows the required data set in the
previous section, but induces experimental biases inherent to the direct shear test. How-
ever, testing is never ideal and therefore not completely corresponding to the wishes of the
operator.
93
2.6.1 Operating mode
Taking into account the collection of expected information, the test procedure consisted of
5 consecutive phases:
• 1st Phase: the application of normal stress since the neutral state until σn cm
• 2nd Phase: the application of shear loading until τcm
• 3rd Phase: the application of N cycles of small amplitude of controlled shear stress
∆τ
• 4th Phase: one great cycle of shear (after N cycles were reached) to failure
• 5th Phase: discharge in shear stress (τ ) and then the normal stress until σn = 0,
respectively.
Figure 2.44 also illustrates the scheme of the test procedure. The 2nd and 3rd phases were
automatically controlled by the system. Additionally, the 4th phase was carried out in order
to characterize the change of interface resistance (δpeak ) due to cyclic loading. This phase
could not be carried out in case of ηcm close to ηpeak due to the early termination (stress
state tended towards critical state).
W Gp
4th phase
Kp 3rd phase Kcm

1 1
2nd phase
1st phase
5th phase Vn cm Vn
1
-K p
-Gp
Figure 2.44: Procedure of cyclic interface direct shear tests
2.6.2 Experimental bias
The rough plates used in this study have roughness amplitude (Rmax ) of about 0.2 mm. So
that the upper half shear box did not rub directly on the rough plate, a spacing of 0.3 mm,
created by a pair of brass foils, was arranged between the plate and the 21 box during the
94
construction of the sand sample to the desired density. Furthermore, this gap between the
shear box and rough plate has to be taken into account because the leakage of sand from
shear box would take place (Figure 2.45). Additional verification of this gap was that a
continuous measurement of the level of the box was also recorded for all purposes by using
LVDT (see Figure 2.45(b)). Thus a certain ”leakage of sample” took place during the test
between the 12 box and the rough plate (Figure 2.46), equivalent to contraction interference.
This loss of material, very small (approximately, 0.2 - 2 g depending on ID0 , σn , ηcm ), was
measured at end of tests. However, this leakage also occurred during the 4th phase which
led to the gradual leakage of sand as increasing the shear displacement [w]. Consequently,
cyclic CNL tests without 4th phase were carried out to specifically isolate the leakage of
sand due to the cycles. From experimental point of view, three possible assumptions to
affect the leakage of sand were;
measuring vertical
movement of sear box
(a) (b)
Figure 2.45: (a) Leakage of sample during cyclic interface shear tests; (b) Additional veri-
fication of vertical movement of shear box
• proportional to time or number of cycles during the tests.

• proportional to accumulated [w]
• proportional to [w] and the vertical displacement of the box.
However, the effect of the vertical relative displacement of the box compared to the rough
plate was difficult to quantify. Figure 2.47 illustrates two possibilities of fictitious contrac-
tions for correcting the normal displacement [u]. The first one which was proportional to
the number of cycles or time led to too much fictitious contraction, especially, in case of
cyclic test with final shear phase (final great cycle). When taking into account the nor-
mal relative displacement measured and the fictitious contraction, the normal displacement
refined became too much dilative. This would be inaccurate in some cases of dense sand.
Whereas the test without final shear phase would be reasonalbe when taking into account
the fictitious contraction proportional to the number of cycles or time. The second one
was proportional to the accumulation of tangential displacement of cyclic loading phase
including the post-cyclic phase (final great cycle). This method provided less fictitious con-
95
traction than the previous one and the refined normal displacement seemed proper for any
conditions. Therefore, the reasonable solution found in this study was that the leakage of
sand was proportional to the accumulated [w]. A simple inclusive correction can then be
performed by considering the loss of sand as the fictitious contraction [u]loss :
mloss g [w]cm Nf
[u]loss = − · (2.18)
γA [w]acc
where:
mloss : total mass of lost sand measured at the end of test;
γ : average specific gravity of sand at the interface during the test;
A : the interface area;
g : gravity acceleration;
[w]cm Nf : mean cyclic shear displacement [w] at the final cycle;
[w]acc : accumulated shear displacement [w], taking into account the post-cyclic phase (final
great cycle);
(a) (b)
Figure 2.46: Examples of loss of material after testing with σn = 120 kPa on loose sand
(ID0 ≈ 30%); (a) loss of material after testing with post-cyclic phase; (b) loss of material
after testing without post-cyclic phase
Consequently, correction of [u]cm was systematically applied to unrefined results [u]unrefined

as:
[w]cm
[u]cm = [u]cm unrefined − [u]loss · (2.19)
[w]acc
96
Figure 2.48 shows the typical [u]cm corrected from the assumption of lost sand proportional
to the accumulated [w] on dense sand with σn = 120 kPa and ηcm ≈ 21 ηpeak (45 < τ < 55
kPa). This example also illustrates the reproducibility of cyclic CNL interface shear tests
by the results of two tests of which the first one included the post-cyclic phase and the
other excluded the post-cyclic phase. In the same way, the shear displacement [w]cm are
also presented. The reproducibility of cyclic CNL interface shear tests was satisfactory.
0.00
fictitious contraction [u]fic (mm)
-0.05
-0.10
ID0 y 90% ; UW = 10 kPa ; Vn = 120 kPa

-0.15 Kcm y 1/2Kpeak ; rough plate ; CNL tests
proportional to [w]; with post-cyclic phase
-0.20 proportional to N or t; with post-cyclic phase
proportional to [w]; without post-cyclic phase
proportional to N or t; without post-cyclic phase
-0.25
1 10 100 1000 10000
N (-)
Figure 2.47: Example of fictitious contractance proportional to [w] and N or t for cyclic
CNL test on dense sand (ID0 ≈ 90%) with σn = 120 kPa, ηcm ≈ 21 ηpeak , ∆τ = 10 kPa
0.00 0.30
ID0 y 90% ; UW = 10 kPa ; Vn = 120 kPa
Kcm y 1/2Kpeak ; rough plate ; CNL tests
0.25
-0.05 10DM120 ; with post-cyclic phase
10DM120 ; without post-cyclic phase
0.20
[u]cm (mm)
[w]cm (mm)
-0.10
ID0 y 90% ; UW = 10 kPa ; Vn = 120 kPa 0.15
-0.15
10DM120_corrected; with post-cyclic phase 0.10
10DM120_corrected; without post-cyclic phase
-0.20
10DM120_unrefined; with post-cyclic phase 0.05
10DM120_unrefined; without post-cyclic phase
-0.25 0.00
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
(a) (b)
Figure 2.48: Cyclic CNL test on dense sand (ID0 ≈ 90%) with σn = 120 kPa, ηcm ≈ 21 ηpeak ,
∆τ = 10 kPa (a) Correction of [u]cm proportional to [w] as a function of N ; (b) Evolution
of [w]cm as a function of N
97
2.7 Identification tests (cyclic CNL tests on rough plate)
A test program was conceived according to the philosophy of the experimental design. These
tests were classified into two groups for two purposes:
• Identification series aimed to study the interface behaviour sujected to cyclic loading
under CNL conditions. The influence of initial normal stress (σn ), amplitude of cycles
(∆τ ), level of mean cyclic stress ratio (ηcm ) on the intensity of displacement was
observed.
• Validation series aimed to highlight the phenomena of interface under CNS condition
and by changing the consecutive package of cycles.
The latter will be described in the following section.
The main parameters of cyclic CNL tests in this study are the initial density of sand (ID0 ),
the level of mean cyclic normal stress (σn cm ), the level of mean cyclic shear stress (τn cm )
and the cyclic amplitude (∆τ ). In terms of normal stress level, the cyclic tests were guided
by the sizes of the current piles as well as by the prescribed limits of the direct shear device.
The small normal stress levels were forbidden for the reason of precision of the tests. In
addition, the same shear box (φ = 60 mm) was used for the whole series of tests. In this
study, the minimum normal stress level was 60 kPa while the maximum normal stress level
was 310 kPa which corresponds to a depth of about 40 m for a bored pile and to 15-20 m
for a driven pile.
Having known the peak (ηpeak ) and characteristic (ηcar ) stress ratios from monotonic CNL
tests, three levels of mean cyclic stress ratio (ηcm ) were decided to perform, i.e. low (ηcm =
0), medium (ηcm ≈ 21 ηpeak ) and high (as close as possible to ηpeak ; ηcm ≈ 9/10ηpeak ). These
tests allowed to get the overview on the cyclic interface response. In case of high level of
ηcm (ηcm ≈ 9/10ηpeak ), ηcm varied in the range 0.50 < ηcm < 0.73 depending on σn , ID0
and cyclic amplitude (∆τ ). Moreover, four supplementary tests with ∆τ = 40 kPa for σn
= 310 kPa were performed in order to properly describe the influence of cyclic amplitude
(at ηcm = 12 ηpeak and ηcm ≈ 9/10ηpeak ). All tests were planned to 104 cycles, unless early
termination was reached. This early termination was found in case of ηcm close to ηpeak .
These tests are summarized in Table 2.4.
The designation of these cyclic CNL tests is as follows:
• The first two digits of the test (e.g. 10, 20 and 40) indicate the amplitude of cycles
in terms of shear stress (∆τ ).
• The first capital letter denotes the type of initial density, ”L” for loose sand and ”D”
for dense sand.
• The following capital letter denotes the level of mean cyclic stress ratio (ηcm ): ”L”
(low, ηcm = 0), ”M” (medium, ηcm ≈ 21 ηpeak ) and ”H” (high, ηcm ≈ 9/10ηpeak )
• The last two or three digits of the test (e.g. 60, 120 and 310) indicate the initial
normal stress (σn ).
98
Table 2.4: Tests for identification, CNL path, parameters
N◦ ID0 σn level of ηcm τ ηcm ∆τ
(%) (kPa) (-) (kPa) (-) (kPa)
10LL60 30 60 0 -5 < τ < 5 0 10

10LM60 30 60 ≈ 1/2ηpeak 10 < τ < 20 0.25 10
10LH60 30 60 ≈ 9/10ηpeak 25 < τ < 35 0.50 10
10LL120 30 120 0 -5 < τ < 5 0 10
10LM120 30 120 ≈ 1/2ηpeak 30 < τ < 40 0.29 10
10LH120 30 120 ≈ 9/10ηpeak 55 < τ < 65 0.50 10
10LL310 30 310 0 -5 < τ < 5 0 10
10LM310 30 310 ≈ 1/2ηpeak 83 < τ < 93 0.28 10
10LH310 30 310 ≈ 9/10ηpeak 170 < τ < 180 0.57 10
10DL60 90 60 0 -5 < τ < 5 0 10
10DM60 90 60 ≈ 1/2ηpeak 15 < τ < 25 0.33 10
10DH60 90 60 ≈ 9/10ηpeak 32 < τ < 42 0.62 10
10DL120 90 120 0 -5 < τ < 5 0 10
10DM120 90 120 ≈ 1/2ηpeak 45 < τ < 55 0.40 10
10DH120 90 120 ≈ 9/10ηpeak 85 < τ < 95 0.75 10
10DL310 90 310 0 -5 < τ < 5 0 10
10DM310 90 310 ≈ 1/2ηpeak 100 < τ < 110 0.34 10
10DH310 90 310 ≈ 9/10ηpeak 50 < τ < 60 0.71 10
20LL60 30 60 0 -10 < τ < 10 0 20
20LM60 30 60 ≈ 1/2ηpeak 5 < τ < 25 0.25 20
20LH60 30 60 ≈ 9/10ηpeak 15 < τ < 25 0.42 20
20LM120 30 120 ≈ 1/2ηpeak 25 < τ < 45 0.29 20
20LL310 30 310 0 -10 < τ < 10 0 20
20LM310 30 310 ≈ 1/2ηpeak 78 < τ < 98 0.28 20
20LH310 30 310 ≈ 9/10ηpeak 160 < τ < 180 0.55 20
20DL60 90 60 0 -10 < τ < 10 0 20
20DM60 90 60 ≈ 1/2ηpeak 10 < τ < 30 0.33 20
20DH60 90 60 ≈ 9/10ηpeak 25 < τ < 45 0.58 20
20DM120 90 120 ≈ 1/2ηpeak 35 < τ < 55 0.37 20
20DL310 90 310 0 -10 < τ < 10 0 20
20DM310 90 310 ≈ 1/2ηpeak 95 < τ < 115 0.34 20
20DH310 90 310 ≈ 9/10ηpeak 210 < τ < 230 0.71 20
40LM310 30 310 ≈ 1/2ηpeak 68 < τ < 108 0.28 40
40LH310 30 310 ≈ 9/10ηpeak 140 < τ < 180 0.52 40
40DM310 90 310 ≈ 1/2ηpeak 85 < τ < 125 0.34 40
40DH310 90 310 ≈ 9/10ηpeak 190 < τ < 230 0.68 40
99
W (kPa) Kpeak Kcm ~ 9/10Kpeak
Kcar
'W = 40 kPa
'W = 20 kPa
'W = 10 kPa
Kcm ~ 1/2Kpeak
Vn (kPa)
60 120 310
Figure 2.49: Tests of identification series with CNL path on dense sand (ID0 ≈ 90%), ∆τ
= 10, 20 and 40 kPa
Figure 2.49 also shows the scheme of identification tests on both densities (ID0 ≈ 90%) with
the amplitude of cycles ∆τ = 10, 20 and 40 kPa.
From the results obtained during this campaign, the first analysis concerns the global be-
haviour of the sample. With a large number of cycles, according to Wichtmann [2005] and
Wichtmann et al. [2005], the fist (irregular) cycle and the subsequent (regular) cycles are
distinguished due to the significant difference in displacements at the beginning of cyclic
loading. Figure 2.50 and 2.51 show respectively the typical τ − [w] as well as [u] − [w]
diagrams and selected number of cycles of cyclic CNL test on loose and dense samples with
σn = 120 kPa, at ηcm = 0 , 21 ηpeak and 9/10ηpeak , ∆τ = 10 kPa.
From typical test results on both densities, it was clear that the first cycle obviously differed
from the subsequent ones. On loose sand the contractive behaviour was observed through
the test. The intensity of relative displacements obviously depended on the level of ηcm .
On dense sand, the existence of characteristic state, passage from contractive to dilative
behaviour, proposed by Luong [1980] was observed when ηcm was in the range of ηcar <
ηcm < ηpeak , otherwise the interface behaved contractively in the range of ηcm < ηcar . This
characteristic state obviously depended on the level of stress ratio (ηcm ) as can be seen in
”10DH120” test. The shear displacement at low level of stress ratio (ηcm = 0) was hardly
observed. At high level of ηcm on both densities, after reaching certain number of cycles
with large enough shear displacement [w], the shear stress could not evolve the next cycle
even though the normal stress was kept constant (the shear stress tended to the critical
state), consequently, the evolution of normal displacement [u] was almost constant (critical
state). It should be noticed that [u] and [u]cm are physically limited due to the mandatory
contact between the grains, which is not the case for [w] and [w]cm
In this study the mean paths during cycles were considered more than the detail of each
cycle. The mean cyclic displacements [u]cm and [w]cm were then taken into account from
N = 2. Due to the amplitudes of cycles in terms of shear stress applied (∆τ ) were small,
100
whether ∆τ = 10, 20, 40 kPa, which led to some ambiguities of corresponding response (see
Figure 2.6). Therefore, certain of the consecutive cycles would be more representative. In
the following figures, [u]cm and [w]cm in dependence of the number of cycles (N ) represent
the mean cyclic path of ten consecutive cycles (e.g., N = 2, 12, 25 and 50... represent
respectively the mean path of N = 2-11, 12-21, 25-34 and 50-59...).
From the experimental observations performed in the identification series and those from the
complete literatures, it is possible to distinguish the main features of the intrinsic behaviour
and the factors influencing the overall behaviour of granular soil-structure interface under
cyclic loading.
120 0.1
ID0 | 30% ; Vn = 120 kPa ; ID0 | 30% ; Vn = 120 kPa ;
100 'W = 10 kPa ; CNL tests 0.0 'W = 10 kPa ; CNL tests
10LH120
80 -0.1
[u] (mm)
60 -0.2
W (kPa)
10LM120 N = 1-100
10LH120 N = 1-100
40 -0.3 N = 200-210
N = 200-210
10LM120 N = 500-510
N = 500-510
20 -0.4 N = 1000-1010
N = 1000-1010
10LL120 N = 2000-2010
N = 2000-2010
0 N = 5000-5010 -0.5 N = 5000-5010
10LL120 N = 9990-10000 N = 9990-10000
-20 -0.6
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
(a) (b)
Figure 2.50: Typical cyclic CNL tests on loose sand (ID0 ≈ 30%) with σn = 120 kPa, ∆τ
= 10 kPa
120 0.3
ID0 | 90% ; Vn = 120 kPa ; ID0 | 90% ; Vn = 120 kPa ;
'W = 10 kPa ; CNL tests 'W = 10 kPa ; CNL tests
100 0.2
80
0.1
10DH120
60
[u] (mm)
W (kPa)
10DH120
0
40 N = 1-100 N = 1-100
10DM120 N = 200-210 N = 200-210
N = 500-510 -0.1 N = 500-510
20 N = 1000-1010 10DM120 N = 1000-1010
N = 2000-2010
-0.2 N = 2000-2010
0 N = 5000-5010 10DL120
N = 9990-10000 N = 5000-5010
10DL120 N = 9990-10000
-20 -0.3
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
[w] (mm) [w] (mm)
(a) (b)
Figure 2.51: Typical cyclic CNL tests on dense sand (ID0 ≈ 90%) with σn = 120 kPa, ∆τ
= 10 kPa
101
2.7.1 Influence of density (ID0 )
The influence of density was observed on two distinct densities (ID0 ≈ 30% and ID0 ≈
90%). Under cyclic CNL condition, the structure of sand-structure interface varied as a
result of the progressive densification due to cycles. For instance, Figure 2.52 and 2.53
show respectively the mean cyclic normal [u]cm and shear [w]cm displacements with the
number of cycles during the cycles N > 1 on loose and dense samples with σn = 120 kPa,
three levels of mean cyclic stress ratio, ηcm (i.e. 0, ≈ 12 ηpeak and ≈ 9/10ηpeak ) with a cyclic
amplitude of ∆τ = 10 kPa. At low density the interface behaved contractively for all levels
of ηcm . The reduction rate in contraction increased as a function of N . On the other hand,
for high density, a dilative behaviour occurred at ηcm close to ηpeak in the range between
ηcar and ηpeak whereas the interface behaved contractively in the range between 0 < ηcm <
ηcar .
Considering the evolution of [w]cm in dependence on the number of cycles (N ), these two
figures qualitatively provide the same information for loose and dense sands. There was an
exponential acceleration of [w]cm when ηcm tended towards ηpeak . At the level of ηcm close
to ηpeak , the interface underwent the significant increase of [w]cm and then the stress state
could not evolve the next cycle due to the experimental limit of [w]cm .
0.0 2.5
ID0 y 30% ; UW = 10 kPa ;
Vn = 120 kPa ; rough plate ; CNL tests
2.0
10LL120
-0.1 10LM120
1.5
[w]cm (mm)
10LH120
[u]cm (mm)
-0.2 1.0
ID0 y 30% ; UW = 10 kPa ;
Vn = 120 kPa ; rough plate ; CNL tests 0.5
-0.3
10LL120 0.0
10LM120
10LH120
-0.4 -0.5
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
(a) (b)
Figure 2.52: Evolution of [u]cm and [w]cm in dependence on the number of cycles (N ) for
cyclic CNL tests on loose sand (ID0 ≈ 30%), rough plate, σn = 120 kPa, ∆τ = 10 kPa
In case of ηcm = 0 (10LL120 and 10DL120), [w]cm was observed to be very small. This was
generally found in the range of ± 0.10 - 0.16 mm (depending on σn and ID0 ). Therefore,
it would be concluded that the interface behaviour had no prominent shear displacement
([w]cm ) at ηcm = 0. In modeling part which will be described in the following chapter [w]cm
is deduced to be zero for ηcm = 0. With increasing ID0 , the rate of [u]cm and [w]cm obviously
decreased.
102
0.20 1.4
ID0 y 90% ; UW = 10 kPa ;
ID0 y 90% ; UW = 10 kPa ;
0.15 1.2 Vn = 120 kPa ; rough plate ; CNL tests
10DL120
10DL120
0.10 10DM120 1.0
[u]cm (mm)
10DM120
[w]cm (mm)
10DH120
0.05 0.8 10DH120
0.00 0.6
-0.05 0.4
-0.10 0.2
-0.15 0.0
-0.20 -0.2
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
(a) (b)
cyclic CNL tests on dense sand (ID0 ≈ 90%), rough plate, σn = 120 kPa, ∆τ = 10 kPa
Figure 2.54 and 2.55 show the evolution of the local tangent stiffness (ks ) and normal
stiffnesses (kn ), corresponding to ηcm , on loose and dense sands (with σn = 120 kPa, ∆τ =
10 kPa ), respectively. The evolution of both kinds of stiffness resulted from cyclic loading
application. Generally, the shear as well as the normal stiffnesses of interface evolving
during the cyclic loading can be expressed as:
dτ dσ
ks = ; kn = (2.20)
d[w] d[u]
However, with the stiffness relation expressed above, it seemed ambiguous in evaluating the
evolution of stiffness during the consecutive cycles. This was due to the ambiguous response
when applying the cyclic shear stress as mentioned before. This ambiguity, especially the
shear stiffness ks , would obviously arise from the tests performed at ηcm = 0. At this level
(ηcm = 0) the evolution of shear displacement was not steady during cyclic shearing or it
could be negligible. To properly achieve this purpose, the evolution of stiffness relavent
to the density would be considered. Therefore, with the small amplitude of cycles and
the mean cyclic path was considered more than the detail of each cycle in this study, the
evolution of stiffness during cyclic loading can therefore be defined in terms of the evolution
of density:
n n
γd γd
ks = ks mono · ; kn = kn mono · (2.21)
γd0 γd0
where the subscript ”mono” denotes the stiffness derived from monotonic tests, γd and γd0
are respectively the current and initial density. The power ”n” is supposed to be constant,
in this study n = 0.50 is used for both kinds of stiffness.
The evolution of normal stiffness (kn ) was obviously higher than that of shear stiffness (ks ).
This was due to the interface thickness being taken into account for evaluating the normal
103
stiffness. For both kinds of stiffness, a stiffening of the interface for the low values of ηcm
(ηcm = 0 and ≈ 21 ηpeak ) could be observed on both densities. The stiffening of the interface
increased as a function of ηcm within the range of ηcm < ηcar . On the other hand a violent
softening, the consequence of the dilative behaviour when ηcm tended towards ηpeak , was
significant on dense sample.
1.6 1.4
ID0 | 30% ; Vn = 120 kPa ; ID0 | 30% ; Vn = 120 kPa ;
10LL120 (cyclic phase) 10LL120 (cyclic phase)
1.5 1.3
10LM120 (cyclic phase)
kn [x104] (kPa/mm)
10LM120 (cyclic phase)
ks [x103] (kPa/mm)
10LH120 (cyclic phase) 10LH120 (cyclic phase)
1.4 1.2
1.3 1.1
1.2 1
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
(a) (b)
Figure 2.54: Evolution of local stiffness in dependence on the number of cycles (N ) for
cyclic CNL tests on loose sand (ID ≈ 30%), rough plate, σn = 120 kPa, ∆τ = 10 kPa; (a)
tangent stiffness (ks ); (b) normal stiffness (kn )
2.6 2.6
ID0 | 90% ; Vn = 120 kPa ; ID0 | 90% ; Vn = 120 kPa ;
2.5
2.5
kn [x104] (kPa/mm)
10DL120 (cyclic phase) 10DL120 (cyclic phase)

ks [x103] (kPa/mm)
10DM120 (cyclic phase)

2.4 10DM120 (cyclic phase)
10DH120 (cyclic phase)
10DH120 (cyclic phase) 2.4
2.3
2.3
2.2
2.2
2.1
2 2.1
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
(a) (b)
Figure 2.55: Evolution of local stiffness in dependence on the number of cycles (N ) for
cyclic CNL tests on dense sand (ID0 ≈ 90%), rough plate, σn = 120 kPa, ∆τ = 10 kPa; (a)
tangent stiffness (ks ); (b) normal stiffness (kn )
104
300 300
ID0 | 30% ; Vn = 310 kPa ; ID0 | 90% ; Vn = 310 kPa ;
250 'W = 10 kPa ; CNL tests 250 'W = 10 kPa ; CNL tests
200 200
10DH310
W (kPa)
150 150
W (kPa)
10LH310
100 100
10DM310
10LM310
50 50
0 0
10LL310 10DL310
-50 -50
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
[w] (mm) [w] (mm)
0 0.2
ID0 | 30% ; Vn = 310 kPa ; ID0 | 90% ; Vn = 310 kPa ;
'W = 10 kPa ; CNL tests 0.15 'W = 10 kPa ; CNL tests
-0.1
0.1
[u] (mm)
[u] (mm)
10LH310
0.05
-0.2 10DH310
0
10LM310
10LL310
-0.05
-0.3 10DM310
-0.1
10DL310
-0.4 -0.15
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
[w] (mm) [w] (mm)
Figure 2.56: Typical cyclic CNL tests with rough plate, σn = 310 kPa, ∆τ = 10 kPa, on
loose sand (ID0 ≈ 30%) and dense sand (ID0 ≈ 90%)
2.7.2 Influence of initial normal stress (σn )
The dependence of the intensity of [u]cm and [w]cm on the initial normal stress was studied
in three different values of σn (60, 120 and 310 kPa). Figure 2.56 shows typical cyclic CNL
tests with σn = 310 kPa, ∆τ = 10 kPa on both densities. Similarly, the overview response
of these tests showed the similar trend as can be seen in the tests with σn = 120 kPa. The
existence of characteristic state could be observed when ηcm was close to ηpeak on dense
sand.
Figure 2.57 shows the evolution of [u]cm and [w]cm in dependence on N (at level of ηcm ≈
1
2 ηpeak ) with three values of initial normal stress (σn = 60, 120 and 310 kPa), cyclic ampli-
tude ∆τ = 10 kPa on both densities. With high value of σn the interface was restrained
to expand during cyclic shearing, therefore the rate of [u]cm and [w]cm in case of σn = 310
kPa obviously decreased on both densities.
105
0.0 0.00
-0.05
-0.2
[u]cm (mm)
ID0 y 30% ; UW = 10 kPa ;
[u]cm (mm)
Kcm y 1/2Kpeak ; rough plate ; CNL tests -0.10

-0.4 ID0 y 90% ; UW = 10 kPa ;
10LM60
10LM120
-0.15
10DM60
10LM310 10DM120
-0.6 10DM310
-0.20
-0.8 -0.25
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
2.0 0.80
ID0 y 30% ; UW = 10 kPa ; ID0 y 90% ; UW = 10 kPa ;
Kcm y 1/2Kpeak ; rough plate ; CNL tests Kcm y 1/2Kpeak ; rough plate ; CNL tests
1.6
10LM60 0.60 10DM60
10LM120 10DM120
[w]cm (mm)
[w]cm (mm)
10LM310 10DM310
1.2
0.40
0.8
0.20
0.4
0.0 0.00
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
cyclic CNL tests on loose sand (ID0 ≈ 30%) and dense sand (ID0 ≈ 90%) with σn = 60,
120 and 310 kPa, ηcm ≈ 21 ηpeak , ∆τ = 10 kPa
In case of ηcm close to ηpeak , Figure 2.58 also shows the evolution of [u]cm and [w]cm as
a function of number of cycles (N ) for three values of initial normal stress (σn ) at ηcm ≈
9/10ηpeak , ∆τ = 10 kPa. On loose sand, the contractive behaviour was evident. It was
found that a number of cycles for reaching the critical state (the state that the cyclic loading
could not evolve the next one) mainly depended on the stress state (e.g. ηcm and ∆τ in τ −σ
plane). Although the level of ηcm = 0.50 (25 < τ < 35 kPa) in case of 10LH60 was slighly
lower than ηcm = 0.57 (170 < τ < 180 kPa) in case of 10LH310, but the ampltitude of cycles
normalized with normal stress as expressed by ∆η = ∆τ /σn would have more influence for
reaching the critical state in case of 10LH60 than 10LH310. A number of cycles for reaching
the critical state in 10LH60 (324 cycles) were less than those in 10LH310 (3,230 cycles).
On dense sand, the dilative behaviour was evident which the rate of dilation decreased with
increasing σn . The intensity rate of [w]cm at this level decreased when σn and ID0 increased.
At low density as well as low normal stress, the interface could reach farther [w]cm (e.g.
[w]cm > 6mm for σn = 60 kPa on ID0 = 30%).
106
0.0 0.20
ID0 y 90% ; UW = 10 kPa ;
0.15
-0.1 10DH60
10DH120
[u]cm (mm)
10DH310
[u]cm (mm)
0.10
-0.2
ID0 y 30% ; UW = 10 kPa ; 0.05
-0.3
10LH60 0.00
10LH120
10LH310
-0.4 -0.05
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
10 10
ID0 y 30% ; UW = 10 kPa ; ID0 y 90% ; UW = 10 kPa ;
8 10LH60 8 10DH60
10LH120 10DH120
[w]cm (mm)
[w]cm (mm)
10LH310 10DH310
6 6
4 4
2 2
0 0
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
cyclic CNL tests on loose sand (ID0 ≈ 30%) and dense sand (ID0 ≈ 90%) with σn = 60,
120 and 310 kPa, ηcm ≈ 9/10ηpeak , ∆τ = 10 kPa
2.7.3 Influence of mean cyclic stress ratio ηcm
The effect of mean cyclic stress ratio is typically presented by the tests of σn = 120 kPa
for three levels of ηcm (i.e. 0, ≈ 21 ηpeak and ≈ 9/10ηpeak ) with a cyclic amplitude ∆τ = 10
kPa (see also Figure 2.52 and 2.53). In case of ηcm close to ηpeak , ηcm was chosen according
to σn , ID0 and cyclic amplitude ∆τ (e.g. ηcm = 0.73 on dense sand and ηcm = 0.50 on
loose sand were chosen for σn = 120 kPa) as shown in Figure 2.59. The intensity of relative
displacements significantly depended on the level of ηcm .
For loose sand, see Figure 2.52, the rate of [u]cm and [w]cm obviously increased with increas-
ing the level of ηcm . However, for the level of ηcm close to ηpeak (ηcm = 0.50) even below ηcar
(ηcar = 0.555), the interface could reach the critical state with a certain number of cycles.
This was due to the significant increase of [w]cm as a function of N . With a significant
increase of [w]cm as a function of N , [w]cm evolved until exceeding the limit in which the
cyclic loading could not evolve further. On loose sand, the interface behaved contractively
and could undergo further [w]cm than that on dense sand.
107
In case of dense sand, see Figure 2.53, within the domain of ηcm < ηcar , the intensity of [u]cm
and [w]cm also increased with increasing the level of ηcm unless ηcm was close to ηpeak . When
ηcm was in the range between ηcar and ηpeak , contrary to those tests mentioned above, the
dilative behaviour took place and consequently the interface could reach the critical state
with a certain number of cycles.
300
UW = 10 kPa
UW = 20 kPa
UW = 40 kPa
200 Kpeak = 0.61

1
W (kPa)
Kcar = 0.555
100 1
0
0 100 200 300 400
Vn (kPa)
(a)
300
UW = 10 kPa
UW = 20 kPa
UW = 40 kPa
200
Kpeak = 0.79
W (kPa)
100 Kcar = 0.555

1
0
0 100 200 300 400
Vn (kPa)
(b)
Figure 2.59: Cyclic CNL tests at ηcm ≈ 9/10ηpeak , depending on σn and ID0 , ∆τ = 10 kPa;
(a) on loose sand (ID0 ≈ 30%) ; (b) on dense sand (ID0 ≈ 90%)
108
0.0 0.2
ID0 y 90% ; UW = 10, 20 kPa ;
0.1
-0.2
[u]cm (mm)
[u]cm (mm)
ID0 y 30% ; UW = 10, 20 kPa ; 0.0
-0.4
10LL60
10LM60 -0.1
10LH60 10DL60
20LL60 10DM60
-0.6 10DH60
20LM60 -0.2
20LH60 20DL60
20DM60
20DH60
-0.8 -0.3
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
10 7
ID0 y 30% ; UW = 10, 20 kPa ; ID0 y 90% ; UW = 10, 20 kPa ;
Vn = 60 kPa ; rough plate ; CNL tests 6 Vn = 60 kPa ; rough plate ; CNL tests
8
10LL60 10DL60
10LM60
5 10DM60
6 10LH60 10DH60
[w]cm (mm)
[w]cm (mm)
4
20LL60 20DL60
4 20LM60 3 20DM60
20LH60 20DH60
2
2
1
0
0
-2 -1
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
Figure 2.60: Influence of cyclic amplitude (∆τ = 10 and 20 kPa) on intensity of [u]cm and
[w]cm in dependence on the number of cycles (N ) for cyclic CNL tests on loose sand (ID0 ≈
30%) and dense sand (ID0 ≈ 90%) with σn = 60 kPa
2.7.4 Influence of cyclic amplitude ∆τ
The tests on the effect of cyclic amplitude were also performed, ∆τ = 10 and 20 kPa, for
three values of initial normal stress. The influence of cyclic amplitude with σn = 60 kPa on
loose and dense samples is first presented. Figure 2.60 shows the intensity of [u]cm and [w]cm
in dependence of N on both densities. For loose sand as for dense sand at ηcm = 0 and ηcm
≈ 21 ηpeak , the intensity of [u]cm and [w]cm increased with increasing the cyclic amplitude.
The rate of the intensity decreases as a function of N . The interface underwent the gradual
contraction with an increasing function of cyclic amplitude within the range of ηcm < ηcar .
At ηcm = 0, although the amplitude of ∆τ = 20 kPa was applied, the mean cyclic path
of [w]cm was still almost zero. In case of ηcm close to ηpeak , on loose sand the contraction
still existed whereas on dense sand the dilative behaviour occurred. It was found that the
dilation of the smaller amplitude (10DH60) was more cosiderable than that of the bigger
one. A number of cycles for reaching the critical state decreased with increasing the cyclic
amplitude. In case of smaller amplitude (∆τ = 10 kPa) the interface could undergo farther
109
than the bigger one (∆τ = 20 kPa), e.g. [w]cm ≈ 6.8 mm for 10DH60.
Figure 2.61 shows the same trend of the influence of the cyclic amplitude on both densities
with the initial normal stress σn = 310 kPa. On loose sand, within the range of ηcm < ηcar
the rate of [u]cm and [w]cm was obviously less than that observed with σn = 60 kPa. At
ηcm close to ηpeak , with ∆τ = 20 kPa (20LH310, 160 < τ < 180) this test could reach the
critical state with 443 cycles whereas in case of ∆τ = 10 kPa (10LH310, 170 < τ < 180) the
interface could evolve more than 3,000 cycles. This could be said that even though the stress
ratio ηcm of 10LH310 was more closer to ηpeak than that of 20LH310, the interface could
undergo farther with smaller amplitude. On dense sand, the influence of cyclic amplitude
induced an increase of the intensity of [u]cm and [w]cm in the range of ηcm < ηcar . In case of
ηcm close to ηpeak the dilative behaviour also occurred. The smaller amplitude (10DH310)
provided more considerable dilation than the bigger one (20DH310).
0.0 0.12
ID0 y 90% ; UW = 10, 20 kPa ;
0.08
-0.1
0.04
[u]cm (mm)
[u]cm (mm)
ID0 y 30% ; UW = 10, 20 kPa ;

0.00
-0.2
10LL310 -0.04 10DL310
10LM310
10LH310 10DM310
20LL310 -0.08 10DH310
-0.3
20LM310 20DL310
20LH310 -0.12 20DM310
20DH310
-0.4 -0.16
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
3.0 1.2
ID0 y 30% ; UW = 10, 20 kPa ; ID0 y 90% ; UW = 10, 20 kPa ;
Vn = 310 kPa ; rough plate ; CNL tests Vn = 310 kPa ; rough plate ; CNL tests
2.5 1.0
10LL310 10DL310
10LM310 10DM310
2.0 0.8
10LH310 10DH310
[w]cm (mm)
[w]cm (mm)
20LL310 20DL310
1.5 0.6 20DM310
20LM310
20LH310 20DH310
1.0 0.4
0.5 0.2
0.0 0.0
-0.5 -0.2
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
[w]cm in dependence on the number of cycles (N ) for cyclic CNL tests on loose sand (ID0 ≈
30%) and dense sand (ID0 ≈ 90%) with σn = 310 kPa
For the purpose of describing properly on amplitude-dependence, the supplementary tests

with ∆τ = 40 kPa were performed for σn = 310 kPa at the level of ηcm ≈ 21 ηpeak and ηcm
≈ 9/10ηpeak . Again, Figure 2.62 shows the evolution of [u]cm and [w]cm in dependence of
110
N for σn = 310 kPa on both densities. Within the range of ηcm < ηcar , the rate of [u]cm
and [w]cm increased with increasing the cyclic amplitude. It is also interesting to consider
the influence of cyclic ampltitude when ηcm close to ηcm . First, considering the increase of
cyclic amplitude on loose sand, with ∆τ = 40 kPa (40LH310, 140 < τ < 180), it could
be seen that ηcm of this test was farther away from ηpeak than those with ∆τ = 10 and
20 kPa (see Figure 2.59(a)). Each of these tests ηmax was kept constant while increasing
the amplitude of cycles, ηcm then decreased. Since ηcm of 40LH310 was lower than that of
20LH310 (∆τ = 20 kPa), the interface could then undergo farther. This could be attributed
to the influence of ηcm . In case of 40LH310, ηcm was in the contractive zone. However,
it could be deduced that with increasing cyclic amplitude a number of cycles for reaching
the critical state decreased. This became evident when the increase of cyclic amplitude was
applied on dense sand. At the level of ηcm close to ηpeak , ηcm of these three tests were in
the range of ηcar < ηcm < ηpeak (see also Figure 2.59(b)). A number of cycles for reaching
the critical state therefore decreased with increasing the cyclic amplitude. The dilative
behaviour which was found in the cases of ηcm close to ηpeak was considerable when the test
was performed with the smaller amplitude.
0.0 0.15
ID0 y 90% ; UW = 10, 20, 40 kPa ;
0.10
-0.1 0.05
[u]cm (mm)
[u]cm (mm)
0.00
-0.2 ID0 y 30% ; UW = 10, 20, 40 kPa ;
Vn = 310 kPa ; rough plate ; CNL tests -0.05
10LM310 10DM310
20LM310 -0.10 20DM310
-0.3 40LM310 40DM310
10LH310 10DH310
20LH310 -0.15 20DH310
40LH310 40DH310
-0.4 -0.20
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
3.0 1.2
ID0 y 30% ; UW = 10, 20, 40 kPa ; ID0 y 90% ; UW = 10, 20, 40 kPa ;
Vn = 310 kPa ; rough plate ; CNL tests Vn = 310 kPa ; rough plate ; CNL tests
2.5 1.0
10LM310 10DM310
20LM310 20DM310
2.0 0.8
[w]cm (mm)
40LM310 40DM310
[w]cm (mm)
10LH310 10DH310
1.5 20LH310 0.6 20DH310
40LH310 40DH310
1.0 0.4
0.5 0.2
0.0 0.0
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
[w]cm in dependence on the number of cycles (N ) for cyclic CNL tests on loose sand (ID ≈
30%) and dense sand (ID ≈ 90%) with σn = 310 kPa, ηcm ≈ 12 ηpeak and ηcm ≈ 9/10ηpeak
111
2.7.5 Post-cyclic interface behaviour
One of the most important phases of the cyclic tests in the identification series is the post-
cyclic phase. This phase can be carried out by one single large cycle after the cyclic phase
(4th phase in Figure 2.44) in order to characterize the change of interface resistance (δpeak )
due to cyclic loading. This post-cyclic phase could not be carried out for ηcm close to ηpeak
due to the early termination. As can be dedeuced from experimental observations, within
the range of ηcm < ηcar the principal characteristic of interface subjected to cyclic loading
under CNL condition was the progressive contraction.
Figure 2.63 shows the typical τ /σ − [w], [u] − [w] and σn − [u] diagrams of cyclic interface
direct shear test for σn = 120 kPa, ηcm ≈ 12 ηpeak , ∆τ = 10 kPa on loose (10LM 120) and
dense (10DM 120) sands after cyclic loading phase. On loose sand, within the range of ηcm
< ηcar , after cyclic phase the interface became denser and consequently the peak stress ratio
(ηpeak ) of cyclic test was slightly higher than that of monotonic test. The interface showed
slightly dilative behaviour at the post-cyclic phase. On dense sand as long as ηcm < ηcar ,
the interface behaved contractively the peak stress ratio after cyclic loading phase was not
much different from monotonic test. At post-cyclic phase, the interface exhibited dilative
behaviour as a result of gradual densification. In comparison with the monotonic tests, the
normal stiffness, kn (slope of ∆σ/∆[u] in σn − [u] diagrams) of cyclic tests after discharging
σn slighly increased. This was due to the cyclic loading phase inducing more densification
on both densities.
Figure 2.64 shows the post-cyclic behaviour for σn = 60 kPa on loose sand with ∆τ = 10
and 20 kPa. As can be observed, the values of peak stress ratio (ηpeak ) of both amplitudes
after cyclic phase were slightly higher than that of monotonic test. Considering the volu-
metric behaviour, [u] − [w] diagrams in Figure 2.64, loose sand which became denser after
undergoing cyclic loading exhibited variations of [u] (slightly dilative behaviour). With in-
creasing the level of ηcm and ∆τ , the peak stress ratio (ηpeak ) increased slightly as a result
of more progressive densification. Nevertheless, in case of ∆τ = 20 kPa at ηcm ≈ 12 ηpeak
(20LM 60) which provided more progressive contraction, ηpeak was lower than those of the
other ones. This might be attributed to either the evolution of grain breakage or the high
evolution of [w]cm during cyclic phase exceeding [w] corresponding to peak stress ratio of
monotonic test. Uesugi et al. [1989] concluded that one of the main factors in the degra-
dation of maximum shear stress was the magnitude of sliding displacement of interface.
Fakharian & Evgin [1997] also explained that under cyclic loading with a sufficient increase
in sliding displacement or slip at interface, the shear stress which mobilized to the peak
value decreased to a residual stress. Considering the specific energy (see Figure 2.65), the
specific energy of grain breakage during cyclic phase at ηcm ≈ 12 ηpeak with ∆τ = 20 kPa
was higher than those of the other ones in cyclic tests, this would lead to the reduction rate
of peak stress ratio. However, this specific energy was still small, the peak stress ratio at
post-cyclic phase of this case was so slightly higher than that of monotonic test.
112
0.8 1
ID0 | 30% ; Vn = 120 kPa ; ID0 I|D |90%
90% ;; V
Vnn ==120
120kPa
kPa; ;
monotonic
'W = 10 kPa ; CNL test
monotonic 'W 'W
= 10 = 10 kPa;;CNL
kPa CNL tests
test
0.8
0.6
K WVn
K WVn (-)
0.6
post-cyclic phase ; 10LM120
0.4 post-cyclic phase ; 10DM120 N = 1-100
N = 1-100
0.4 N = 200-210
N = 200-210
N = 500-510 N = 500-510
0.2 N = 1000-1010 N = 1000-1010
N = 2000-2010 0.2 N = 2000-2010
N = 5000-5010 N = 5000-5010
N = 9990-10000 N = 9990-10000
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0 0.3
ID0 | 90% ; Vn = 120 kPa ;
-0.1 0.2 'W = 10 kPa ; CNL test
monotonic
-0.2 0.1
monotonic
-0.3 0.0
[u] (mm)
[u] (mm)
-0.4 -0.1
post-cyclic phase ; 10LM120 N = 1-100 N = 1-100
-0.5 N = 200-210 -0.2 N = 200-210
N = 500-510 post-cyclic phase ; 10DM120 N = 500-510
-0.6 N = 1000-1010 -0.3 N = 1000-1010
ID0 | 30% ; Vn = 120 kPa ; N = 2000-2010 N = 2000-2010
-0.7 'W = 10 kPa ; CNL test N = 5000-5010 -0.4 N = 5000-5010
N = 9990-10000 N = 9990-10000
-0.8 -0.5
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
160 160
ID0 | 30% ; Vn = 120 kPa ; ID0ID|| 90%
90% ; ;VV
n=n= 120
120 kPakPa
; ;
140 shearing phase 'W = 10 kPa ; CNL test 140 shearing phase 'W'W==1010 kPa
kPa ;;CNL
CNLtests
test
120 120
100 100
Vn (kPa)
Vn (kPa)
monotonic
80 monotonic test
80 N = 1-100
N = 1-100 discharge Vn N = 200-210
60 N = 200-210 60 N = 500-510
discharge Vn N = 500-510 discharge Vn N = 1000-1010
40 N = 1000-1010 40 N = 2000-2010
N = 2000-2010 N = 5000-5010
N = 5000-5010
applying Vn
N = 9990-10000
20 N = 9990-10000 20 post-cyclic phase
post-cyclic phase applying Vn
0 0
-0.60 -0.50 -0.40 -0.30 -0.20 -0.10 0.00 0.10 0.20 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20
[u] (mm) [u] (mm)
Figure 2.63: Post-cyclic phase for cyclic CNL tests with rough plate, σn = 120 kPa, ηcm ≈
1
2 ηpeak , ∆τ = 10 kPa on loose (ID0 ≈ 30%) and dense (ID0 ≈ 90%) sands
113
1 1
ID0 | 30% ; Vn = 60 kPa ; ID0 | 30% ; Vn = 60 kPa ;
0.8 0.8
0.6 0.6
K WVn
K WVn
0.4 0.4

0.2 0.2
monotonic
monotonic
0 10LL60 (cyclic phase) 0 20LL60 (cyclic phase)
10LM60 (cyclic phase) 20LM60 (cyclic phase)
-0.2 10LL60 (post-cyclic phase) -0.2 20LL60 (post-cyclic phase)
10LM60 (post-cyclic phase) 20LM60 (post-cyclic phase)
-0.4 -0.4
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0 0.0
-0.2
-0.2
-0.4
-0.4
[u] (mm)
[u] (mm)
-0.6

-0.6
monotonic -0.8 monotonic
-0.8 10LM60 (cyclic phase) 20LM60 (cyclic phase)
ID0 | 30% ; Vn = 60 kPa ; 10LL60 (post-cyclic phase)
-1.0 ID0 | 30% ; Vn = 60 kPa ; 20LL60 (post-cyclic phase)
'W = 10 kPa ; CNL tests 10LM60 (post-cyclic phase) 'W = 20 kPa ; CNL tests 20LM60 (post-cyclic phase)
-1 -1.2
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.64: Post-cyclic phase for cyclic CNL tests with rough plate, σn = 60 kPa, ηcm =
0 and ≈ 12 ηpeak , ∆τ = 10 and 20 kPa on loose sand (ID0 ≈ 30%)
1000 1000
ID0 | 30% ; Vn = 60 kPa ; ID0 | 30% ; Vn = 60 kPa ;
[W] (Joules/m2)
[W] (Joules/m2)
100 100

10 monotonic 10 monotonic
10LL60 (post-cyclic phase) 20LL60 (post-cyclic phase)
1 1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
(a) (b)
Figure 2.65: Evolution of specific energy during cyclic loading (CNL tests) with rough plate,
σn = 60 kPa, ηcm = 0 and ≈ 12 ηpeak on loose sand (ID0 ≈ 30%); (a) ∆τ = 10 kPa; (b) ∆τ
= 20 kPa
114
1 1
ID |
ID0 90% ;;VVnn==6060kPa
| 90% kPa; ;
'W
'W==10
10kPa
kPa ; CNL
CNLtests
tests 0.8
0.8
0.6
0.6
K WVn
K WVn
0.4
0.4 D|
IID0 |90%
90%; ;VV
n= 60 kPa ;
n = 60 kPa ;

0.2

'W == 20
'W 20 kPa
kPa;;CNL
CNLtests
tests
monotonic
0.2 monotonic
10DL60 (cyclic phase)
0 20DL60 (cyclic phase)
10DL60 (post-cyclic phase) 20DM60 (cyclic phase)
0 -0.2
10DM60 (post-cyclic phase) 20DL60 (post-cyclic phase)
20DM60 (post-cyclic phase)
-0.2 -0.4
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.3 0.3
0.2 0.2
0.1 0.1
[u] (mm)
0 0
[u] (mm)
-0.1 -0.1
ID0 | 90% ; Vn = 60 kPa ; | 90% ; Vn =6060
D | 90% ; Vn =
IID0 kPakPa
; ;
'W = 10 kPa ; CNL tests 'W
'W == 20 kPa
kPa;;CNL
CNLtests
tests
-0.2 -0.2
monotonic monotonic
-0.3 10DL60 (cyclic phase) -0.3 20DL60 (cyclic phase)
10DM60 (cyclic phase) 20DM60 (cyclic phase)
-0.4 10DL60 (post-cyclic phase) -0.4 20DL60 (post-cyclic phase)
10DM60 (post-cyclic phase) 20DM60 (post-cyclic phase)
-0.5 -0.5
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.66: Post-cyclic phase for cyclic CNL tests with rough plate, σn = 60 kPa, ηcm =
0 and ≈ 12 ηpeak , ∆τ = 10 and 20 kPa on dense sand (ID0 ≈ 90%)
1000 1000
100
[W] (Joules/m2)
100
[W] (Joules/m2)
ID0 |ID90%
| 90%; V; nV=n =6060kPa
kPa;;
D|
IID0 |90%
90%; ;VV
n= 6060
n= kPa ; ;
kPa

'W == 10

'W ='W20= kPa
20 kPa ; CNL
; CNL tests
tests
'W 10 kPa
kPa;;CNL
CNLtests
tests
monotonic
10 monotonic 10
10DL60 (cyclic phase) 20DL60 (cyclic phase)
10DL60 (post-cyclic phase) 20DL60 (post-cyclic phase)
1 1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
(a) (b)
Figure 2.67: Evolution of specific energy during cyclic loading (CNL tests) with σn = 60
kPa, ηcm = 0 and ≈ 12 ηpeak on dense sand (ID0 ≈ 90%); (a) ∆τ = 10 kPa; (b) ∆τ = 20 kPa
115
On dense sand as shown in Figure 2.66, the post-cyclic phase also provided the slight
increase of peak stress ratio (ηpeak ). Within the range of ηcm < ηcar the interface behaved
contractively, the peak stress ratio at this phase seemed to decrease as a function of ηcm
although a gradual densification during cyclic loading was observed. The dilative behaviour
([u] − [w] diagrams) observed at this phase was not so much different from monotonic
test. Figure 2.67 also shows the evolution of specific energy during cyclic phase as well
as post-cyclic phase. At the beginning of post-cyclic phase, the specific energy increased
significantly and was higher than that of monotonic test afterward increased continuously
and then coincided with that of monotonic test.
In case of σn = 310 kPa, Figure 2.68 shows the behaviour of post-cyclic phase on loose
sand which exhibited slight variation of peak stress ratio among those tests. With ∆τ =
10 kPa, in case of ηcm = 0 and ≈ 21 ηpeak , there was not a significant increase of the peak
stress ratio (ηpeak ) at post-cyclic phase even though ηcm ≈ 12 ηpeak (10LM 310) provided
more densification during cyclic phase. Again, at the level of ηcm ≈ 12 ηpeak with increasing
∆τ , (∆τ = 10, 20 and 40 kPa), despite the gradual densification that accompanied cyclic
loading there was not the significant increase of peak stress ratio. Furthermore, the peak
stress ratio of these three tests at post-cyclic phase seemed to be similar to that of monotonic
test. On loose sand, the contractive behaviour during cyclic loading resulted in interface
densification (increasing as a function of ∆τ ) but the significant variation of volumetric
behaviour ([u] − [w] diagrams) could hardly be observed during post-cyclic shearing.
On dense sand as shown in Figure 2.70, with ∆τ = 10 kPa, an increase of ηcm which
practically induced more densification had no variations of peak stress ratio. Moreover, at
ηcm ≈ 21 ηpeak , the peak stress ratio at post-cyclic phase decreased slightly with increasing
∆τ (∆τ = 10, 20 and 40 kPa). Even though an increase of ∆τ induced more gradual
contraction, there was not significant variation of normal displacement [u] (dilation rate)
at post-cyclic phase. This phenomenon might be attributed to the crushing and wear of
the grains resulting from an increase of inter-granular particles in the localized shear zone
between sand and rough plate during cyclic phase (Al-Douri & Poulos [1992], Tabucanon et
al. [1995]). However, this contribution seems to be in contrast with that concluded by Uesugi
& Kishida [1986]. They concluded that during cyclic loading an increase of crushing paricles
due to the high intensity of stress state increasing the normalized the surface roughness (Rn )
of the soil-structure interface then led to the higher coefficient of friction. Nevertheless, they
also concluded that the characteristics of sand (e.g. shape of grains, hardness etc.) and the
sliding displacement at the interface mainly influenced the coefficient of friction.
Considering the specific energy, see Figure 2.69, with high value of σn , ηcm and ∆τ , an
increase of specific energy after cyclic loading phase would lead to gradual crushing and
wear of grains, especially in case of dense sand. Consequently, the peak stress ratio of
interface (ηpeak ) at post-cyclic phase was slightly lower than that of monotonic test.
116
1 1
ID0 | 30% ; Vn = 310 kPa ; ID0 | 30% ; Vn = 310 kPa ; CNL tests
0.8 'W = 10 kPa ; CNL tests
0.8
0.6
K WVn
K WVn
0.6
0.4
monotonic
monotonic 0.4 10LM310 (cyclic phase)
0.2 10LL310 (cyclic phase) 20LM310 (cyclic phase)
0.2 10LM310 (post-cyclic phase)
0 10LL310 (post-cyclic phase) 20LM310 (post-cyclic phase)
-0.2 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0 0.1
ID0 | 30% ; Vn = 310 kPa ; CNL tests
0
-0.1
-0.1
-0.2
-0.2
[u] (mm)
[u] (mm)
ID0 | 30% ; Vn = 310 kPa ;

-0.3 'W = 10 kPa ; CNL tests -0.3
monotonic
monotonic -0.4 10LM310 (cyclic phase)
-0.4
10LL310 (cyclic phase) 20LM310 (cyclic phase)
10LM310 (cyclic phase) -0.5 40LM310 (cyclic phase)
-0.5 10LM310 (post-cyclic phase)
10LL310 (post-cyclic phase)
-0.6 20LM310 (post-cyclic phase)
-0.6 -0.7
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.68: Post-cyclic phase for cyclic CNL tests with rough plate, σn = 310 kPa, ∆τ =
10, 20 and 40 kPa on loose sand (ID0 ≈ 30%)
10000 10000
ID0 | 90% ; Vn = 310 kPa ;
ID0 | 30% ; Vn = 310 kPa ;
'W = 20 kPa ; CNL tests
1000 1000
[W] (Joules/m2)
[W] (Joules/m2)
100 100
monotonic
20LL310 (cyclic phase) monotonic
10 20LM310 (cyclic phase) 10 20DL310 (cyclic phase)
20LL310 (post-cyclic phase)
20DL310 (post-cyclic phase)
20LM310 (post-cyclic phase) 20DM310 (post-cyclic phase)
1 1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
(a) (b)
Figure 2.69: Evolution of specific energy during cyclic loading (CNL tests) with rough plate,
σn = 310 kPa, ηcm = 0 and ≈ 12 ηpeak , ∆τ = 20 kPa; (a) on loose sand (ID0 ≈ 30%); (b) on
dense sand (ID0 ≈ 90%)
117
1 1
0.8
0.8
0.6
K WVn
K WVn (-)
0.6
0.4
0.4 monotonic
monotonic 10DM310 (cyclic phase)
0.2
10DL310 (cyclic phase) 20DM310 (cyclic phase)
0.2 10DM310 (post-cyclic phase)
0 10DL310 (post-cyclic phase)
20DM310 (post-cyclic phase)
-0.2 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.2 0.2
0.1 0.1
[u] (mm)
[u] (mm)
0 0
-0.1 monotonic
-0.1
monotonic
10DL310 (cyclic phase) 40DM310 (cyclic phase)
-0.2 10DM310 (cyclic phase) -0.2 10DM310 (post-cyclic phase)
10DL310 (post-cyclic phase) 20DM310 (post-cyclic phase)
-0.3 -0.3
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.70: Post-cyclic phase for cyclic CNL tests with rough plate, σn = 310 kPa, ∆τ =
10, 20 and 40 kPa on dense sand (ID0 ≈ 90%)
The observation of the grain breakage within the interface was also carried out after finishing
the cyclic tests. As can be seen in monotonic tests, there were many factors that influenced
the grain breakage. Again, the same procedure (sieving method) was still carried out in
order to investigate the influence of cyclic loading on grain breakage. Figure 2.71 and 2.72
show typical gains size distribution curves resulting from monotonic and cyclic interface
shear tests with various conditions of σn = 60 and 310 kPa, respectively.
In comparison with monotonic tests, cyclic tests resulted in more grain breakage within
the localized shear zone between sand and rough surface. However, in cyclic test series the
particle size (D50 ) still did not change so much but this would be considerable when taking
into account the fine particle size (D10 ). At low value of normal stress (σn = 60 kPa),
there was a small variation of fine particle size (D10 ) on dense sand whereas the variation
of paticle size could hardly be observed on loose sand. On dense sand, with increasing ηcm
and ∆τ , the small variation of fine grains (D10 ) could be observed (Figure 2.71(b)). In case
of σn = 310 kPa (see Figure 2.72), the steady shear stress at a high value as well as the
continuous contraction in cyclic CNL tests resulting in an increase in the specific energy
induced an intense grain breakage and the formation of fine particles (resulting in a drop
118
in particle size of the material). Although the specific energy during cyclic phase on loose
sand was slighly higher than that on dense sand, the energy on both densities then evolved
in the same value during post-cyclic phase. Indeed, dense sand has more tendency of grain
breakage within the localized shear zone than loose sand. The abrasion between the grains
and the surface roughness of plate on dense sand therefore provided an increase in crushing
and wear of grains.
100 100
90 90
80 80
finer by weight (%)
finer by weight (%)

70 70
60 60
50 ID0 = 30% ; Vn = 60 kPa 50 ID0 = 90% ; Vn = 60 kPa
initial (laser method)
40 40 initial (laser method)
initial (sieving method)l initial (sieving method)
30 monotonic CNL test 30 monotonic CNL test
10LL60 10DL60
20 10LM60 20 10DM60
10 20LL60 10 20DL60
20LM60 20DM60
0 0
0.01 0.1 1 10 0.01 0.1 1 10
(a) (b)
Figure 2.71: Grain size distribution of sand after testing with various CNL conditions, rough
plate, σn = 60 kPa; (a) on loose sand (ID0 ≈ 30%); (b) on dense sand (ID0 ≈ 90%)
100 100
90 90
80 80
finer by weight (%)
finer by weight (%)
70 70
60 60
ID0 = 30% ; Vn = 310 kPa ID0 = 90% ; Vn = 310 kPa
50 50
initial (laser method) initial (laser method)
40 40
initial (sieving method) initial (sieving method)
30 monotonic CNL test 30 monotonic CNL test
10LL310 10DL310
20 10LM310 20 10DM310
20LM310 20DM310
10 10
40LM310 40DM310
0 0
0.01 0.1 1 10 0.01 0.1 1 10
(a) (b)
Figure 2.72: Grain size distribution of sand after testing with various CNL conditions, rough
plate, σn = 310 kPa; (a) on loose sand (ID0 ≈ 30%); (b) on dense sand (ID0 ≈ 90%)
The change in particle size described by D10 (a shift of grading curve to the left) was
greater for increasing ID0 , σn , ηcm and ∆τ . The small amplitude of cycle (∆τ = 10 kPa)
can even lead to the crushing and wear of the sand particles if the interface undergoes a
large number of cycles (as can be seen in this study, N = 104 ). This may contribute to the
119
small degradation of the peak stress ratio at post-cyclic phase. The fine grains resulting
from crushing grains during cyclic loading then replaced the void within the interface zone.
Consequently, this led to further contraction of interface.
2.8 Validation tests (cyclic CNS tests on rough plate)
In this study, some validation tests were performed in order to investigate the interface
behaviour subjected to cyclic loading under constant normal stiffness (CNS) condition.
This would be more realistic to study the behaviour of habitual problems often encontered
in practice such as pile foundations, soil nailing etc. The first series of validation tests was
to test the cyclic pseudo-creep CNS path with three values of normal stiffness (k) and initial
normal stress (σn0 ). The second series was intended to test the memory parameters of the
constitutive law, which made it possible to connect the cycles of different characteristics.
These tests were performed by changing the amplitude of cycles in succession. The detail
of the tests in this series is summarized in Table 2.5
2.8.1 Influence of normal stiffness (k)
In case of cyclic CNS tests, all tests were performed with σn0 = 60, 100 and 310 kPa on
both densities while the stress ratio (ηcm0 ) varied according to ID0 and the normal stiffness
(k). In this section, three values of normal stiffness k = 1000, 2000 and 5000 kPa/mm
are focused. From our experience, an interface behaviour with an imposed normal stiffness
of 5000 kPa/mm is not far from the ”constant volume” condition at least for the current
normal stress levels (< 500 kPa). In addition, ever from our experience, the normal stiffness
corresponding to the current piles, cast in situ as well as driven, is much less than 5000
kPa/mm. The main factors that influence the interface behaviour (i.e. ID0 , σn0 , ηcm0 and
∆τ ) were the same as those presented under CNL condition.
The influence of normal stiffness on the cyclic interface was more significant than that on
monotonic interface behaviour. The principal characteristic of cyclic interfaces under CNS
condition was the gradual degradation of normal stress accompanied with the degradation
of shear stress as a result of the progressive contraction of soil adjacent to the structural
material. Figure 2.73 illustrates the relationship between the variation of normal stress
(∆σn ) and the variation of normal displacement (∆[u]) to represent the imposed normal
stiffness (k) during cyclic loading in case of k = 1000 and 2000 kPa/mm on both densities.
Although there were somewhat many scatters during the test, the trend fitted through the
data indicated that the overall test data correlated well with this relationship. The scatters
were mainly due to the applied normal stress from generating engine in two directions.
Moreover, taking into account the fictitious contraction due to the leakage of sand during
cyclic tests which induced less contraction also led to some of the scatters (as can be seen in
the phase of towards critical state). The important feature during all these cyclic tests was
that the stress ratio (ηcm ) increased (because of the degradation of σn ) from ηcm0 prescribed
at the beginning of the test to the critical value (ηcrit ).
120
Table 2.5: Tests for validation, CNS path, parameters
N◦ ID0 σn k level of ηcm0 τ ηcm ∆τ
(%) (kPa) (kPa/mm) (-) (kPa) (-) (kPa)
10LM60 1000k 30 60 1000 ≈ 1/2ηpeak 13 < τ < 23 0.30 10

10LM100 1000k 30 100 1000 ≈ 1/2ηpeak 25 < τ < 35 0.30 10
20LM100 1000k 30 100 1000 ≈ 1/2ηpeak 20 < τ < 40 0.30 20
10LM100 2000k 30 100 2000 ≈ 1/3ηpeak 15 < τ < 25 0.20 10
10LL100 5000k 30 100 5000 ≈ 1/4ηpeak 10 < τ < 20 0.15 10
20LL100 5000k 30 100 5000 ≈ 1/4ηpeak 5 < τ < 25 0.15 20
10LM310 1000k 30 310 1000 ≈ 1/2ηpeak 88 < τ < 98 0.30 10
10DM60 1000k 90 60 1000 ≈ 1/2ηpeak 16 < τ < 26 0.35 10
10DH60 1000k 90 60 1000 ≈ 1/2ηpeak 20 < τ < 30 0.42 10
10DM100 1000k 90 100 1000 ≈ 1/2ηpeak 30 < τ < 40 0.35 10
10DH100 1000k 90 100 1000 ≈ 2/3ηpeak 45 < τ < 55 0.50 10
20DM100 1000k 90 100 1000 ≈ 1/2ηpeak 25 < τ < 45 0.35 20
10DM100 2000k 90 100 2000 ≈ 1/2ηpeak 30 < τ < 40 0.35 10
10DH100 2000k 90 100 2000 ≈ 2/3ηpeak 45 < τ < 55 0.50 10
10DM100 5000k 90 100 5000 ≈ 1/2ηpeak 30 < τ < 40 0.35 10
10DH100 5000k 90 100 5000 ≈ 2/3ηpeak 45 < τ < 55 0.50 10
20DM100 5000k 90 100 5000 ≈ 1/2ηpeak 25 < τ < 45 0.35 20
10DM310 1000k 90 310 1000 ≈ 1/2ηpeak 105 < τ < 115 0.35 10
10DH310 1000k 90 310 1000 ≈ 2/3ηpeak 150 < τ < 160 0.50 10
10 20LM120 30 120 CNL ≈ 1/2ηpeak 30 < τ < 40 0.29 10
25 < τ < 45 0.29 20
20 10LM120 30 120 CNL ≈ 1/2ηpeak 25 < τ < 45 0.29 20
30 < τ < 40 0.29 10
10 20DM120 30 120 CNL ≈ 1/2ηpeak 45 < τ < 55 0.40 10
40 < τ < 60 0.40 20
20 10DM120 30 120 CNL ≈ 1/2ηpeak 35 < τ < 55 0.37 20
40 < τ < 50 0.37 10
121
140 120
ID0 | 90% ; Vn0 = 100 kPa ; Kcm0 = 0.50 ID0 | 30% ; Vn0 = 100 kPa ; Kcm0 = 0.20
k = 1000 kPa/mm ; 45 < W < 55 kPa k = 2000 kPa/mm ; 15 < W < 25 kPa
120 100
N = 1-100 N = 1-2
N = 190-200 N=3
100
N = 500-510 80 N=4
Vn (kPa)
N = 1000-1010 N=5
Vn (kPa)
80 N = 2000-2010 towards critical state

N = 3000-3010
60
60 N = 3980-3990
towards critical state 40
40
20 20
k 'Vn'[u] = 2000 kPa/mm
0 0
-0.14 -0.12 -0.10 -0.08 -0.06 -0.04 -0.02 0.00 -0.1 -0.08 -0.06 -0.04 -0.02 0 0.02
[u] (mm) [u] (mm)
(a) (b)
Figure 2.73: Relationship of imposed stiffness during cyclic CNS tests, with rough plate,
σn0 = 100 kPa, ∆τ = 10 kPa ; (a) k = 1000 kPa/mm, ηcm0 = 0.50 on dense sand ID0 ≈
90%; (b) k = 2000 kPa/mm, ηcm0 = 0.20 on loose sand ID0 ≈ 30%
On loose sand, as can be seen in the preliminary series of monotonic tests, the significant
degradation of normal stress associated with shear stress was observed as a result of con-
traction during shearing. In cyclic condition, the significant degradation of normal stress
(σn ) as a result of more progressive contraction accompanied with N led to the critical
state easily. An increase of normal stiffness on loose sand induced the stress state to ac-
celerate towards the critical state. Figure 2.74 shows a comparison of stress paths and the
degradation of σn during cyclic CNS tests with the initial normal stress σn0 = 100 kPa, ∆τ
= 10 kPa on loose sand. The first one was performed with k = 1000 kPa/mm at ηcm0 =
0.30 while the second one was performed with k = 5000 kPa/mm at ηcm0 = 0.15. As can
be observed on τ − σn and σn − [w] planes, the influence of k was evident on loose sand.
Althoughh the value of ηcm0 was small (ηcm0 = 0.15) there was a significant degradation of
σn with k = 5000 kPa/mm. In this case, in spite of small [w] the normal stress decreased
rapidly during the first two cycles (from 100 to 20 kPa) until the 6th cycle, while the mean
shear stress was kept constant (τcm = 15 kPa), after that the stress state could not evolve
to the next cycle and consequently moved towards the critical state line.
A comparison between the distinction of σn0 (σn0 = 60 and 310 kPa) was also performed
at the same level of ηcm0 (ηcm0 = 0.30), k = 1000 kPa/mm on loose sand. The test with
σn0 = 60 kPa was performed at 13 < τ < 23 kPa (Figure 2.75) while the case of σn0 = 310
kPa was performed at 88 < τ < 98 kPa (Figure 2.76). Obviously, with low initial normal
stress the stress state could undergo only two cycles whereas the latter the stress state could
undego more than 2,000 cycles. The degradation rate of normal stress with σn0 = 60 kPa
increased rapidly during the first two cycles and then the stress state could not evolve to
the next one. This indicates that whenever the interface subjected to cyclic loading under
CNS condition with low initial normal stress on loose sand, it can easily reach the critical
state due to the initial stress state close to the critical state line.
122
100 100
80 80
N = 1-2
N = 1-5 N = 3-4
N = 6-10
W (kPa)
W (kPa)
60 60 N = 5-6
N = 11-15 Kpeak = 0.61 Kpeak = 0.61
towards critical state
towards critical state 1 1
Kcrit = 0.566
40 40 Kcrit = 0.566
1
1
20 20
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Vn (kPa) Vn (kPa)
120 120
100 100
N = 1-5 N = 1-2
N = 6-10 N = 3-4
80 80 N = 5-6
N = 11-15
Vn (kPa)
Vn (kPa)
towards critical state towards critical state

60 60
40 40
20 20
0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5
[w] (mm) [w] (mm)
0 0.02
0
-0.02 N = 1-2
N = 1-5
N = 3-4
N = 6-10 -0.02 N = 5-6
-0.04 N = 11-15
[u] (mm)

[u] (mm)

-0.04
-0.06
-0.06
-0.08
-0.08
-0.1 -0.1
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5
[w] (mm) [w] (mm)
Figure 2.74: Cyclic CNS tests on loose sand (ID0 ≈ 30%) with rough plate, σn0 = 100 kPa,
∆τ = 10 kPa; (left) k = 1000 kPa/mm, ηcm0 = 0.30; (right) k = 5000 kPa/mm, ηcm0 =
0.15
123
60 80
k = 1000 kPa/mm ; 13 < W < 23 kPa 70 k = 1000 kPa/mm ; 13 < W < 23 kPa
50 N=1
N=1 60 N=2
N=2 towards critical state
40
towards critical state Kpeak = 0.61 50
Vn (kPa)
W (kPa)
1
30 Kcrit = 0.566 40
1 30
20
20
10
10
0 0
0 10 20 30 40 50 60 70 80 0 1 2 3 4 5 6
Vn (kPa) [w] (mm)
0 80
-0.02 N=1
N=1 60 N=2
N=2 towards critical state
-0.04 50
Vn (kPa)
[u] (mm)
40
-0.06 30
20
-0.08
10 k = UVn/U[u] = 1000 kPa/mm
-0.1 0
0 1 2 3 4 5 6 -0.1 -0.08 -0.06 -0.04 -0.02 0 0.02
[w] (mm) [u] (mm)
k = 1000 kPa/mm, ∆τ = 10 kPa at 13 < τ < 23 kPa
On the other hand, dense sand exhibiting low attitude to contract required a greater number
of cycles as well as the tangential displacement [w] to reach the critical state. Figure 2.77
shows the typical results of cyclic CNS tests on dense sand with σn0 = 100 kPa, ηcm0 = 0.35,
∆τ = 10 kPa, for k = 1000 and 5000 kPa/mm. During cyclic loading the interface showed
continuous contraction which induced the degradation in normal stress. This contraction
continued as a function of number of cycles which associated with the shear displacement
([w]) and then led to more degradation in σn as a result of the interface densification (see
σn − [w] and [u] − [w] planes in Figure 2.77). Subsequently, with large shear displacement
the interface showed more contraction and then the stress state moved towards the critical
line as shown in τ − σn plane. After reaching the critical state the cycles could not evolve
to the next one, the stress state then decreased continuously and coincided the critical state
line.
In case of k = 1000 kPa/mm with ηcm0 = 0.35, while the stress state was in critical condition
the stress path did not coincide the critical line in τ − σn plane. This was due to the
limitation of the shear displacement transducer (LVDT) as mentioned before in the section
of interface direct shear device. Generally, the horizontal relative movement of the plate
124
was set up in the range of ±6.5 mm in accordance with the limitation of LVDT. In this case
the interface could undergo farther (more than 10 mm with N = 11,972) and reached the
limitation of horizontal movement, subsequently the plate was obstructed while the stress
state decreased continuously.
300 400
ID0 | 30% ; Vn0 = 310 kPa ; Kcm0 = 0.30
ID0 | 30% ; Vn0 = 310 kPa ; Kcm0 = 0.30
250 N = 1-50 N = 1-50
N = 100-110 300 N = 100-110
N = 190-200 N = 190-200
200 N = 500-510 Kpeak = 0.61 N = 500-510
250
N = 1000-1010
W (kPa)
Vn (kPa)
N = 1000-1010 1
N = 1500-1510 N = 1500-1510
150 Kcrit = 0.566
200
N = 2000-2010 N = 2000-2010
N = 2025-2035 1 N = 2025-2035
150 towards critical state
100
50
50
0 0
0 50 100 150 200 250 300 350 400 0 2 4 6 8 10 12
Vn (kPa) [w] (mm)
0.1 400
ID0 | 30% ; Vn0 = 310 kPa ; Kcm0 = 0.30 N = 1-50 N = 1-50 ID0 | 30% ; Vn0 = 310 kPa ; Kcm0 = 0.30
k = 1000 kPa/mm ; 88 < W < 98 kPa N = 100-110 350 N = 100-110 k = 1000 kPa/mm ; 88 < W < 98 kPa
N = 190-200 N = 190-200
0 N = 500-510
300 N = 500-510
N = 1000-1010 N = 1000-1010
N = 1500-1510 N = 1500-1510
-0.1 N = 2000-2010 250
Vn (kPa)
N = 2000-2010
[u] (mm)
N = 2025-2035
200 N = 2025-2035
-0.2 150
100
-0.3
k 'Vn'[u] = 1000 kPa/mm
50
-0.4 0
0 2 4 6 8 10 12 -0.4 -0.3 -0.2 -0.1 0 0.1
[w] (mm) [u] (mm)
k = 1000 kPa/mm, ∆τ = 10 kPa at 88 < τ < 98 kPa
To describe the evolution of the mean cyclic stress state (represented by ηcm = τcm /σn cm )
tending towards the critical state, the ηcm − [w] and ηcm − N diagrams of the tests with
σn0 = 100 kPa as shown in Figure 2.78 are then prensented. During the interface subjected
to cyclic loading, the shear stress was kept constant while the normal stress decreased as a
function of N as well as [w] then the mean cyclic stress ratio ηcm which started from the
beginning (ηcm0 = 0.35) increased and then moved to the critical state line. An increase of
normal stiffness induced more degradation of normal stress, consequently the mean cyclic
stress ratio ηcm increased significantly. Obviously, a number of cycles for reaching the
critical state with high value of k (k = 5000 kPa/mm) were less than those with the other
ones. However, in case of k = 2000 and 5000 kPa/mm the interface could almost undergo
the same value of shear displacement ([w]) but difference in N for reaching the critical state
line. Unlike those tests in case of k = 1000 kPa/mm the interface could undergo farther
125
with a large number of cycles.
100 120
N = 1-100 ID0 | 90% ; Vn0 = 100 kPa ; Kcm0 = 0.35 ID0 | 90% ; Vn0 = 100 kPa ; Kcm0 = 0.35
N = 190-200
N = 500-510 100
80
N = 1000-1010 N = 1-100
N = 2000-2020 Kpeak = 0.79 N = 190-200
80 N = 500-510 Kpeak = 0.79
N = 5000-5010 1
60
W (kPa)
1
W (kPa)
N = 10000-10010 N = 1000-1010
N = 11960-11972 Kcrit = 0.566 N = 1500-1510
60 Kcrit = 0.566
towards critical state 1 N = 1650-1662
40 towards critical state 1
40
20
20
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120 140
Vn (kPa) Vn (kPa)
120 140
100 120
N = 1-100
N = 190-200
100 N = 500-510
80 N = 1000-1010
Vn (kPa)
Vn (kPa)
80 N = 1500-1510
N = 1650-1662
60 N = 1-100 towards critical state
N = 190-200 60
N = 500-510
40 N = 1000-1010
N = 2000-2020 40
N = 5000-5010
20 N = 10000-10010 20
N = 11960-11972
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w] (mm) [w] (mm)
0.02 0.04
0 0.02
-0.02
0
-0.04
[u] (mm)
[u] (mm)
N = 1-100
N = 190-200 -0.02
-0.06 N = 500-510 N = 1-100
N = 1000-1010 N = 190-200
-0.04 N = 500-510
-0.08 N = 2000-2020
N = 5000-5010 N = 1000-1010
N = 10000-10010 -0.06 N = 1500-1510
-0.1 N = 11960-11972 N = 1650-1662
-0.12 -0.08
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w] (mm) [w] (mm)
Figure 2.77: Cyclic CNS tests on dense sand (ID0 ≈ 90%) with rough plate, σn0 = 100 kPa,
∆τ = 10 kPa; (left) k = 1000 kPa/mm, ηcm0 = 0.35; (right) k = 5000 kPa/mm, ηcm0 =
0.35
126
1 1
Kpeak = 0.79 Kpeak = 0.79

0.8 0.8
Kcm = Wcm/Vn cm (-)
Kcm = Wcm/Vn cm (-)

0.6 Kcrit = 0.566 0.6 Kcrit = 0.566
0.4 0.4
ID0 | 90% ; CNS tests ; ID0 | 90% ; CNS tests ;
Kcm0 = 0.35 ; Vn = 100 kPa Kcm0 = 0.35 ; Vn = 100 kPa
0.2 k = 1000 kPa/mm 0.2 = 1000 kPa/mm
k1000k
1000k
k = 2000 kPa/mm
2000k k = 2000 kPa/mm
2000k
k = 5000 kPa/mm
5000k k = 5000 kPa/mm
5000k
0 0
0 2 4 6 8 10 12 0 1 10 100 1000 10000 100000
[w]cm (mm) N (-)
(a) (b)
Figure 2.78: Evolution of ηcm as a function of N and [w]cm of cyclic CNS tests, σn0 = 100
kPa on dense sand (ID0 ≈ 90%)
In case of k = 5000 kPa/mm which may be attributed to very high value of stiffness (approx-
imately constant volume condition), the volumetric behaviour of interface represented by
[u] − [w] diagrams in Figure 2.74 and 2.77 showed the contractive behaviour which seemed
to be constant during cyclic loading on both densities.
2.8.2 Influence of ηcm0
The series of validation also covered the influence of ηcm0 in CNS condition. First, this
observation is typically represented by the tests carried out with σn0 = 100 kPa, ∆τ = 10
kPa, k = 2000 kPa/mm on two levels of ηcm0 (ηcm0 = 0.35 and 0.50). Figure 2.79 illustrates
how the level of ηcm0 influenced the interface behaviour under cyclic CNS condition. An
increase of ηcm0 obviously resulted in less number of cycles for reaching the critical condition.
However, the degradation rate of normal stress accompanied with [w] at ηcm0 = 0.50 was
less than that at ηcm0 = 0.35 (see σ − [w] diagram). During N = 1500 - 2000 the normal
stress of ηcm0 = 0.50 slightly increased as a result of more densification and then decerased
as a function of [w] and N . The stress state of high ηcm0 which was close to ηcrit at the
beginning of cyclic loading moved towards critical state line earlier than that of low ηcm0
as can be seen in τ − σn plane.
When considering the influence of ηcm0 with σn0 = 60 kPa, these two cyclic tests were
performed which ηcm0 was not much different. The first one was performed at ηcm0 = 0.35
(16 < τ < 26 kPa) while ηcm0 of the second one was 0.42 (20 < τ < 30 kPa) as shown
in Figure 2.80. A number of cycles for reaching the critical state at ηcm0 = 0.42 kPa were
significantly less than those at ηcm0 = 0.35. In case of ηcm0 = 0.42 the stress state could
reach the critical state and evolved until N = 178 whereas ηcm0 = 0.35 the stress state could
undergo more than 1,000 cycles. As can be seen in τ − σn plane, at the beginning of cyclic
shearing phase, the stress state at ηcm0 = 0.42 was close to the critical state line and the
stress state could move to the critical state line during the first 20 cycles.
127
120 120
k = 2000 kPa/mm ; 30 < W < 40 kPa N = 1-100 k = 2000 kPa/mm ; 45 < W < 55 kPa
100 N = 1-100 100 N = 190-200
N = 190-200 N = 500-510
N = 500-510
Kpeak = 0.79 N = 1000-1010
80 N = 1000-1010 80 N = 1500 - 1510 Kpeak = 0.79
W (kPa)
N = 2000-2010 1
W (kPa)
N = 2000-2010
N = 3000-3101
60 N = 2500-2510
60 N = 4000-4010 Kcrit = 0.566
N = 4850-4858 N = 2540-2550
40 40
20 20
Kcrit = 0.566

0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Vn (kPa) Vn (kPa)
120 140
ID0 | 90% ; Vn0 = 100 kPa ; Kcm0 = 0.35 N = 1-100 ID0 | 90% ; Vn0 = 100 kPa ; Kcm0 = 0.50
100 N = 500-510 120
N = 1000-1010
N = 2000-2010
N = 3000-3101 100
80 N = 4000-4010
Vn (kPa)
Vn (kPa)
N = 4850-4858 80
60 N = 1-100
60 N = 190-200
N = 500-510
40 N = 1000-1010
40 N = 1500 - 1510
N = 2000-2010
20 20 N = 2500-2510
N = 2540-2550
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w] (mm) [w] (mm)
0.02 0.04
0.00 0.02
0
-0.02
[u] (mm)
[u] (mm)
-0.02
-0.04 N = 1-100
N = 1-100 N = 190-200
N = 190-200 -0.04 N = 500-510
N = 500-510 N = 1000-1010
-0.06 N = 1000-1010
-0.06 N = 1500 - 1510
N = 2000-2010 N = 2000-2010
N = 3000-3101 N = 2500-2510
-0.08 N = 4000-4010 -0.08 N = 2540-2550
N = 4850-4858 towards critical state
-0.10 -0.1
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w] (mm) [w] (mm)
∆τ = 10 kPa, k = 2000 kPa/mm, ηcm0 = 0.35 and 0.50
128
70 70
ID0 | 90% ; Vn0 = 60 kPa ; Kcm0 = 0.35 N = 1-20 ID0 | 90% ; Vn0 = 60 kPa ; Kcm0 = 0.42
N = 1-100 k = 1000 kPa/mm ; 16 < W < 26 kPa k = 1000 kPa/mm ; 20 < W < 30 kPa
60 N = 30-40
N = 190-200 60
N = 50-60
N = 500-510 N = 80-90
50 N = 800-810 50 N = 100-110 Kpeak = 0.79
Kpeak = 0.79 N = 120-130
N = 1000-1010
1 N = 150-160 1
W (kPa)
40 N = 1030-1042 40
W (kPa)
N = 170-178
towards critical state K = 0.566
crit towards critical state Kcrit = 0.566
30 1 30 1
20 20
10 10
0 0
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Vn (kPa) Vn (kPa)
80 80
N = 1-100 ID0 | 90% ; Vn0 = 60 kPa ; Kcm0 = 0.42 N = 1-20
ID0 | 90% ; Vn0 = 60 kPa ; Kcm0 = 0.35 N = 30-40
k = 1000 kPa/mm ; 16 < W < 26 kPa
N = 190-200 k = 1000 kPa/mm ; 20 < W < 30 kPa N = 50-60
N = 500-510
N = 800-810 N = 80-90
60 N = 1000-1010 60 N = 100-110
N = 1030-1042 N = 120-130
towards critical state N = 150-160
Vn (kPa)
N = 170-178
Vn (kPa)

40 40
20 20
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w] (mm) [w] (mm)
0.02 0.02
0 0.00
-0.02 -0.02
[u] (mm)
[u] (mm)
-0.04 -0.04
N = 1-20
N = 1-100 N = 30-40
-0.06 N = 190-200 -0.06 N = 50-60
N = 500-510 N = 80-90
N = 800-810 N = 100-110
N = 120-130
-0.08 N = 1000-1010 -0.08 N = 150-160
N = 1030-1042 N = 170-178
-0.1 -0.10
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w] (mm) [w] (mm)
∆τ = 10 kPa, k = 1000 kPa/mm, ηcm0 = 0.35 and 0.42
129
It is also interesting to investigate the influence of ηcm0 on the interface behaviour under
CNS condition with high value of initial normal stress. Figure 2.81 shows the result of cyclic
CNS test with σn0 = 310 kPa, k = 1000 kpa/mm at ηcm0 = 0.50 (150 < τ < 160 kPa).
Although this test was performed with high ηcm0 a number of cycles for reaching the critical
state were more than 17,800 cycles (in comparison with σn0 = 100 kPa at the same level of
ηcm0 which N ≈ 3900). The degradation rate of normal stress increased slowly until [w] ≈
4 mm, subsequently the stress state tended to the critical state.
300 400
ID0 | 90% ; Vn0 = 310 kPa ; Kcm0 = 0.50
ID0 | 90% ; Vn0 = 310 kPa ; Kcm0 = 0.50
k = 1000 kPa/mm ; 150 < W < 160 kPa 350
250 k = 1000 kPa/mm ; 150 < W < 160 kPa
N = 1-100
Kpeak = 0.79 300
N = 190-200
N = 500-510 1
200
N = 1000-1010 250
Vn (kPa)
W (kPa)
N = 2000-2010 N = 1-100
N = 5000-5010 N = 190-200
150 200 N = 500-510
N = 10000-10010
N = 1000-1010
N = 12000-12010 150 N = 2000-2010
100 N = 15000-15010
N = 17870-17876 Kcrit = 0.566 N = 5000-5010
100 N = 10000-10010
towards critical state 1 N = 12000-12010
50 N = 15000-15010
50 N = 17870-17876
0 0
0 50 100 150 200 250 300 350 400 0 2 4 6 8 10 12 14
Vn (kPa) [w] (mm)
(a) (b)
∆τ = 10 kPa, k = 1000 kPa/mm, ηcm0 = 0.50
In case of σn0 = 310 kPa at ηcm0 = 0.35 (105 < τ < 115 kPa), this test was performed
until N = 32669 while the stress state was still so far from the critical line, subsequently
the post-cyclic phase was performed instead (Figure 2.82). In this case, the degradation of
normal stress as a function of N increased very slow. Considering the evolution of mean
cyclic stress ratio ηcm as a function of N , during cyclic phase ηcm increased with a very
slow rate. With N = 32669 the interface could only undergo [w] ≈ 0.30 mm (see Figure
2.84). This can be descriebed that with high σn0 and low level of ηcm0 the initial stress
state was so far from the critical state line, then the stress state was able to evolve further.
Consequently, a large number of cycles were required to reach the critical state. Figure
2.83 also illustrates a comparison between the the degradation of mean cyclic normal stress
(σn cm ) and the evolution of mean cyclic stress ratio (ηcm ) as a function of N of these two
tests. As can be seen, although the degradation rate of σn at ηcm0 = 0.50 was less than
that of ηcm0 = 0.35 the evolution of ηcm of the former evolved close to the critical state
line (ηcrit ) and finally reached ηcrit . This can clearly be described in Figure 2.83(b) when
considering the evolution of ηmax which can be expressed as
∆η
ηcm + 2
ηmax =
1− ηcm ∆η
2
The evolution of ηmax in case of ηcm0 = 0.35 continued without having the trend to reach
130
the critical state line whereas ηmax of the test with ηcm0 = 0.50 could reach this critical
state line.
350 500
N = 1-100
300 N = 190-200
N = 500-510 400
N = 1000-1010
250 N = 2000-2010
N = 5000-5010
N = 10000-10010 300
Vn (kPa)
200
W (kPa)
N = 15000-15010
N = 20000-20010
150 N = 25000-25010
N = 30000-30010 200
N = 32660-32669
100 post-cyclic phase
100
50
Kcrit = 0.566
1
0 0
0 50 100 150 200 250 300 350 400 450 1 10 100 1000 10000 100000
Vn (kPa) N (-)
(a) (b)
∆τ = 10 kPa, k = 1000 kPa/mm, ηcm0 = 0.35
400 1.0
ID0 | 90% ; Vn0 = 310 kPa ;
Kcm0 = 0.35 and 0.50 ; k = 1000 kPa/mm
Kpeak = 0.79
0.8
300
Kcm = Wcm/Vn cm (-)
Kmax
Vn cm (kPa)
Kcrit = 0.566
0.6
200
Kmax
Kcm0 = 0.35
0.4
Kcm0 = 0.50

100 ID0 | 90% ; CNS tests ; Vn0 = 310 kPa

0.2 k = 1000 kPa/mm
'W = 10 kPa ; Kcm0 = 0,35
sn310_sh10_eta=0,35
'W = 10 kPa ; K=0,50
sn310_sh10_eta cm0 = 0,50
0 0.0
0.1 1 10 100 1000 10000 100000 0 1 10 100 1000 10000 100000
N (-) 1
(a) (b)
Figure 2.83: Effect of the level of ηcm0 with σn0 = 310 kPa, ∆τ = 10 kPa, k = 1000
kPa/mm on dense sand (ID ≈ 90%), rough plate; (a) the degradation rate of normal stress
; (b) evolution of ηmax
Considering the post-cyclic phase in case of ηcm0 = 0.35 (σn0 = 310 kPa), this phase
started at σn ≈ 230-240 kPa as shown in Figure 2.84. In comparison with the stress path
that obtained from monotonic CNS test with σn0 = 310 kPa, the peak shear stress (τmax )
at post-cyclic phase was obviously lower than that of monotonic CNS test. However, when
considering the peak stress ratio (ηpeak ), ηpeak after cyclic loading was slightly lower than
those obtained whether from monotonic CNS or CNL tests. Although the cyclic loading
induced more densification of interface, there was not a large difference in dilation between
131
the post-cyclic phase and monotonic CNS test (see [u] − [w] diagram in Figure 2.84). This
is in accordance with Tabucanon et al. [1995]’s investigation. They also reported that there
was a lower stress recovery during post-cyclic response and the loss of strength increased
with increasing the number of cycles due to the smaller volume change accompanying shear
loading of the interface.
350 450
ID0 | 90% ; Vn0 = 310 kPa ; Kcm0 = 0.35
k = 1000 kPa/mm ; 110 < W < 115 kPa monotonic CNS
N = 1-100 400
300 N = 190-200
N = 500-510 350
250 N = 1000-1010
N = 2000-2010 300
N = 5000-5010 N = 1-100
Vn (kPa)
200 N = 10000-10010 N = 190-200
W (kPa)
N = 15000-15010 250 N = 500-510

N = 20000-20010 N = 1000-1010
150 N = 25000-25010 200 N = 2000-2010
N = 30000-30010 N = 5000-5010
N = 32660-32669 150 N = 10000-10010
100 post-cyclic phase N = 15000-15010
monotonic CNS N = 20000-20010
100 N = 25000-25010
50 Kcrit = 0.556 N = 30000-30010
50 ID0 | 90% ; Vn0 = 310 kPa ; Kcm0 = 0.35
N = 32660-32669
k = 1000 kPa/mm ; 110 < W < 115 kPa post-cyclic phase
0 0
0 50 100 150 200 250 300 350 400 450 0 1 2 3 4 5 6
Vn (kPa) [w] (mm)
1 0.1
ID0 | 90% ; Vn0 = 310 kPa ; Kcm0 = 0.35
k = 1000 kPa/mm ; 110 < W < 115 kPa
0.8 0.05
[u] (mm)
monotonic CNS monotonic CNS N = 1-100

K WVn (-)
N = 190-200
0.6 0 N = 500-510
N = 1000-1010
monotonic CNL N = 2000-2010
0.4 -0.05 N = 5000-5010
post-cyclic phase N = 10000-10010
N = 15000-15010
N = 20000-20010
0.2 -0.1 N = 25000-25010
ID0 | 90% ; Vn0 = 310 kPa ; Kcm0 = 0.35 N = 30000-30010
k = 1000 kPa/mm ; 110 < W < 115 kPa N = 32660-32669
post-cyclic phase
0 -0.15
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.84: Comparison between post-cyclic CNS and monotonic CNS tests on dense sand
(ID0 ≈ 90%) with rough plate, σn0 = 310 kPa, ∆τ = 10 kPa, k = 1000 kPa/mm, ηcm0 =
0.35
2.8.3 Influence of cyclic amplitude (∆τ )
Two cyclic amplitudes (∆τ = 10 and 20 kPa) were performed on both densities with k =
1000 and 5000 kPa/mm. Figure 2.85 illustrates the influence of cyclic amplitude (∆τ ) on
dense sand with k = 5000 kPa/mm, σn0 = 100 kPa at ηcm0 = 0.35. As can be conceived
that an increase of ∆τ at the same level of ηcm0 induced the interface trending towards
more progressive densification and consequently the gradual degradation in normal stress.
Therfore, the critical state could be reached with a few number of cycles. With smaller
amplitude of cycle ∆τ , in contrast, a large number of cycles for reaching the critical state
would be required. In comparison between these two tests, with k = 5000 kPa/mm a
132
number of cycles for reaching the critical state were not much different (i.e. N = 1662
with ∆τ = 10 kPa and N = 1176 with ∆τ = 20 kPa). In this study, it was found that
the influence of cyclic amplitude was significant when performing the tests with k = 1000
kPa/mm, σn0 = 100 kPa. In case of ∆τ = 10 kPa at ηcm0 = 0.35 (30 < τ < 40 kPa) as
can be seen before (see Figure 2.77), a number of cycles for reaching the critical state were
11,972 cycles. Figure 2.86 shows the result of cyclic CNS test with ∆τ = 20 kPa (25 < τ
< 45 kPa) which provided the significant degradation in number of cycles for reaching the
critical state (i.e. N = 3630 for reaching ηcrit ). Figure 2.87 can illustrate how the stress
state in case of ∆τ = 20 kPa can reach the critical state early than ∆τ = 10 kPa. Although
the rate of the evolution of ηcm as a function of N in case of ∆τ = 20 kPa was slower
than that of ∆τ = 10 kPa the maximum stress ratio (ηmax ) of a bigger amplitude could
reach the critical state line early. Some of the main factors influencing the cyclic interface
(e.g. σn0 , ηcm0 and ∆τ ) are presented in Figure 2.88 in the form of the evolution of ηcm
as well as the degradation of σn in dependence of N . As can be seen during cyclic CNS
tests, the degradation of σn as a function of N induced an increase in ηcm which evolved
and subsequently moved towards the critical state line.
120 100
100 N = 100-110
80
N = 1-100 Kpeak = 0.79
N = 190-200
N = 190-200
80 N = 500-510 Kpeak = 0.79 N = 500-510 1
W (kPa)
60 N = 1000-1010 Kcrit = 0.566

W (kPa)
N = 1000-1010 1
N = 1500-1510 N = 1170-1176 1
60 Kcrit = 0.566
towards critical state 1 40
40
20
20
0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Vn (kPa) Vn (kPa)
140 120
120 100
N = 1-100
N = 190-200
100 N = 500-510
N = 1000-1010 80
Vn (kPa)
N = 1500-1510
Vn (kPa)
80
N = 1650-1662
towards critical state 60
60
N = 1-50
40 N = 100-110
40 N = 190-200
N = 500-510
20 20 N = 1000-1010
N = 1170-1176
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w] (mm) [w] (mm)
Figure 2.85: Influence of ∆τ on normal stiffness (k = 5000 kPa/mm ) of cyclic CNS tests
with rough plate, σn0 = 100 kPa, ηcm0 = 0.35 on dense sand (ID0 ≈ 90%)
133
100 120
100
80 N = 1-100
N = 190-200
Kpeak = 0.79
N = 500-510 80
Vn (kPa)
60 N = 1000-1010 1
W (kPa)
N = 2000-2010
N = 3000-3010 60 N = 1-100
N = 3600-3630 N = 190-200
40 N = 500-510
40 N = 1000-1010
N = 2000-2010
20 N = 3000-3010
Kcrit = 0.566 20
N = 3600-3630
0 0
0 20 40 60 80 100 120 0 2 4 6 8 10 12 14
Vn (kPa) [w] (mm)
(a) (b)
∆τ = 20 kPa, k = 1000 kPa/mm, ηcm0 = 0.35
Kpeak = 0.79
0.8
Kcm = Wcm/Vn cm (-)
Kcrit = 0.566 Kmax ('W = 20 kPa)

0.6
Kmax ('W = 10 kPa)
0.4
ID0 | 90% ; CNS tests ; Vn0 = 100 kPa

0.2 k = 1000 kPa/mm
'W = 10 kPa ; Kcm0 = 0,35
sh10_eta0,35
'W = 20 kPa ; Kcm0 = 0,35
sh20_eta=0,35
0
0 1 10 100 1000 10000 100000
1
Figure 2.87: Comparison between the evolution of ηcm as a function of N of cyclic CNS
tests with ∆τ = 10 and 20 kPa on dense sand (ID0 ≈ 90%) with rough plate, σn0 = 100
kPa, k = 1000 kPa/mm, ηcm0 = 0.35
134
ID0 | 90% ; CNS tests ;k = 1000 kPa/mm
500 V kPa ; 'W = 10 kPa ; Kcm0 = 0.35
n0 = 60 cm0=0,35
sn60_sh10_eta
1
V kPa ; 'W = 10 kPa ; Kcm0 = 0.42
n0 = 60 cm=0,42
sn60_sh10_eta
V = 100
sn100_sh10_eta kPa ; 'W = 10 kPa ; Kcm0 = 0.35
cm0=0,35
V
n0
; 'W = 20 kPa ; Kcm0 = 0.35 Kpeak = 0.79
400 n0 = 100 kPa
sn100_sh20_eta cm0=0,35
V = 100
sn100_sh10_eta kPa ; 'W = 10 kPa ; Kcm0 = 0.50
cm0=0,50
0.8
n0
V n0 = 310 kPa ; 'W = 10 kPa ; Kcm0 = 0.35
Kcm = Wcm/Vn cm (-)

sn310_sh10_eta cm0=0,35
V n0 = 310 kPa
sn310_sh10_eta ; 'W = 10 kPa ; Kcm0 = 0.50
cm0=0,50
300 0.6 Kcrit = 0.566

Vn cm (kPa)
200 0.4
100 0.2
0 0
0 1 10 100 1000 10000 100000 0 1 10 100 1000 10000 100000
1 1
(a) (b)
Figure 2.88: Influence of σn0 , ηcm0 and ∆τ on cyclic CNS tests with rough plate, dense sand
(ID0 ≈ 90%); (a) degradation of σn as a function of N ; (b) evolution of ηcm as a function
of N
For loose sand the increase of cyclic amplitude led to the significant degradation of normal
stress as a result of gradual contraction. In comparison with the case of σn0 = 100 kPa,
∆τ = 10 kPa as presented in Figure 2.74, an increase of k as well as ∆τ could lead to the
critical state rapidly. In case of ∆τ = 20 kPa as shown in Figure 2.89, both tests (k =
1000 kPa/mm at ηcm0 = 0.30 and k = 5000 kPa/mm at ηcm0 = 0.15) could undergo only
2 cycles. It was found that σn dropped significantly during the first two cycles afterwards
the stress state could not evolve to the next cycle.
From many cyclic tests under CNS condition described above, it can be concluded that the
worse condition is the test being performed with high level of ηcm0 , large-amplitude of cycle,
high value of k, low initial normal stress (σn0 ) on loose sand. This conclusion can clearly be
explained by the evolution of ηcm in dependence of N on loose sand for various conditions
as shown in Figure 2.90. All tests in this figure started from ηcm = 0.30, the test with low
initial normal stress (σn0 = 60 kPa) and the case of σn0 = 100 kPa with ∆τ = 20 kPa, ηcm
moved rapidly towards the critical state line with a few number of cycles.
Considering the grain breakage after testing, Figure 2.91 shows typically the grain breakage
represented by the grain size distribution curves of the tests with σn0 = 310 kPa, ∆τ = 10
kPa, k = 1000 kPa/mm on both densities. It was found that the fine particle (D10 ) after
cyclic CNS loading on dense sand showed the variation of grading curves from the initial
state (the shift of grading curves to the left). With low level of ηcm0 (10DM 310 1000k), the
grain breakage was considerably observed. Because the interface underwent cyclic shearing
for a long period (N ≈ 32600), the grains within the intense shear were able to be crushed.
On loose sand, the grading curve of the cyclic CNS test did not show the significant variation
from the initial state.
135
100 100
80 80 N=1
N=1
N=2 N=2
60 60 Kpeak = 0.61
W (kPa)
W (kPa)
Kpeak = 0.61
1
1
40 40 Kcrit = 0.566
1
20 20
Kcrit = 0.566
1
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120 140
Vn (kPa) Vn (kPa)
120 120
100 N=1 100
N=1
N=2
towards critical state N=2
80 80 towards critical state
Vn (kPa)
Vn (kPa)
60 60
40 40
20 20
0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5
[w] (mm) [w] (mm)
Figure 2.89: Influence of ∆τ and k on cyclic CNS tests with rough plate, σn0 = 100 kPa,
on loose sand (ID0 ≈ 30%)
500 0.8
ID0 | 30% ; CNS tests ; Kcm0 = 0.30
k = 1000 kPa/mm 0.7
Vn0 = 60 kPa ; 'W = 10 kPa
sn60_1000k_sh10_eta=0,30
Kpeak = 0.61
400 Vn0 = 100 kPa ; 'W = 10 kPa
sn100_1000k_sh10_eta=0,30
0.6
Vn0 = 100 kPa ; 'W = 20 kPa
Kcm = Wcm/Vn cm (-)
sn100_1000k_sh20_eta=0,30
Vn0 = 310 kPa ; 'W = 10 kPa
sn310_1000k_sh10_eta=0,30 Kcrit = 0.566
0.5
Vn (kPa)
300
0.4
200 0.3 ID0 | 30% ; CNS tests ; Kcm0 = 0.30
k = 1000 kPa/mm
0.2 Vn0 = 60 kPa ; 'W = 10 kPa
sn60_1000k_sh10_eta=0,30
100 Vn0 = 100 kPa ; 'W = 10 kPa
sn100_1000k_sh10_eta=0,30
0.1 Vn0 = 100 kPa ; 'W = 20 kPa
sn100_1000k_sh20_eta=0,30
Vn0 = 310 kPa ; 'W = 10 kPa
sn310_1000k_sh10_eta=0,30
0 0
0 1 10 100 1000 10000 0 1 10 100 1000 10000
1 1
(a) (b)
Figure 2.90: Influence of σn0 , and ∆τ on cyclic CNS tests with rough plate, dense sand
(ID0 ≈ 30%); (a) degradation of σn as a function of N ; (b) evolution of ηcm as a function
of N
136
100 100
90 90
80 80
finer by weight (%)

finer by weight (%)
70 70
60 60
ID0 = 90% ; Vn = 310 kPa
50 50
ID0 = 30% ; Vn = 310 kPa CNS tests
40 CNS tests 40 initial (laser method)
initial (laser method) initial (sieving method)
30 30
initial (sieving method) monotonic_1000k
20 20 10DH310_1000k
monotonic_1000k
10 10 10DM310_1000k
10LM310_1000k
0 0
0.01 0.1 1 10 0.01 0.1 1 10
(a) (b)
Figure 2.91: Grain size distribution of sand after testing with various CNS conditions, rough
plate, σn = 310 kPa, k = 1000 kPa/mm, ∆τ = 10 kPa ; (a) on loose sand (ID0 ≈ 30%);
(b) on dense sand (ID0 ≈ 90%)
2.9 Validation tests (cyclic CNL tests by changing cyclic am-

plitude)
This series was carried out by changing the sequence and amplitude of two consecutive
packages (∆τ = 10 and 20 kPa), each with 5,000 cycles, in CNL condition with σn = 120
kPa, ηcm ≈ 21 ηpeak . These tests were still in the range of ηcm < ηcar . Figure 2.92 illustrates
the typical τ − [w] and [u] − [w] diagrams of cyclic CNL direct shear interface by changing
the amplitude of cycles on loose sand. These two tests were performed by changing the
package with ∆τ = 10 ka at ηcm = 0.30 (30 < τ < 40 kPa) and ∆τ = 20 ka at ηcm = 0.30
(25 < τ < 45 kPa) vice versa. These diagrams show, for each package, the first 100 cycles
of cyclic loading and 10 consecutive cycles at N= 200, 500, 1000, 2000 and 5000.
Figure 2.92 also illustrates that when the package of cycles with small amplitude (∆τ = 10
kPa) was first applied, the subsequent package with large amplitude (∆τ = 20 kPa) then
induced the reduction of displacement rate. However, the subsequent package with large
amplitude (∆τ = 20 kPa) was applied, the first cycle of this package did not significantly
differ from the following ones as can be observed from the ordinary tests. It can be said
that the precedent package with ∆τ = 10 kPa caused gradual densification and consequently
induced a reduction in rate of [u] and [w].
137
80 80
ID0 | 30% ; Vn = 120 kPa ; ID0 | 30% ; Vn = 120 kPa ;
'W = 10 D 20 kPa ; CNL test 'W = 20 D 10 kPa ; CNL test
60 60
W (kPa)
W(kPa)
40 40
20 20
'W = 10 kPa 'W = 20 kPa 'W = 20 kPa 'W = 10 kPa
W < 40 kPa W < 45 kPa W < 45 kPa W < 40 kPa
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 1.2
[w] (mm)
[w] (mm)
0 0
ID0 | 30% ; Vn = 120 kPa ; ID0 | 30% ; Vn = 120 kPa ;
-0.1 'W = 10 D 20 kPa ; CNL test 'W = 20 D 10 kPa ; CNL test
-0.2 -0.2
[u] (mm)
[u] (mm)
-0.3
-0.4 -0.4
-0.5
-0.6 'W = 10 kPa 'W = 20 kPa -0.6

W < 40 kPa W < 45 kPa 'W = 20 kPa 'W = 10 kPa
-0.7
W < 45 kPa W < 40 kPa
-0.8 -0.8
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 1.2
[w] (mm) [w] (mm)
Figure 2.92: τ − [w] and [u] − [w] diagrams of cyclic CNL tests by changing the amplitude
of cycles, 5000 cycles of each amplitude, with rough plate on loose sand (ID0 ≈ 30%), σn0
= 120 kPa, ηcm ≈ 12 ηpeak
When starting with large amplitude (∆τ = 20 kPa), the displacements, [u] and [w], caused
by this package were more prominent than those with ∆τ = 10 kPa. The displacements
during the subsequent package of ∆τ = 10 kPa evolved hardly. The first package with ∆τ
= 20 kPa resulted in more progressive densification and led to the significant reduction of
displacement rate when the subsequent package of ∆τ = 10 kPa was applied. The total
displacements at N = 10000 of the test starting with ∆τ = 20 kPa were larger than those
with ∆τ = 10 kPa. This was due to the larger displacement in the first package with large
amplitude.
138
0.0 0.00
-0.1 -0.05
-0.2 -0.10
[u]cm (mm)
[u]cm (mm)
-0.3 -0.15
ID0 y 30% ; UW = 10, 20 kPa ; Vn = 120 kPa ID0 y 90% ; UW = 10, 20 kPa ; Vn = 120 kPa
-0.4 -0.20
10LM120 10DM120
20LM120 20DM120
-0.5 10-20LM120 -0.25 10-20MD120
20-10LM120 20-10DM120
-0.6 -0.30
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
1.0 0.5
ID0 y 30% ; UW = 10, 20 kPa ; Vn = 120 kPa ID0 y 90% ; UW = 10, 20 kPa ; Vn = 120 kPa
0.8 0.4
10LM120 10DM120
20LM120 20DM120
[w]cm (mm)
10-20LM120
[w]cm (mm)
0.6 0.3 10-20MD120

20-10LM120 20-10DM120
0.4 0.2
0.2 0.1
0.0 0.0
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
Figure 2.93: Evolution of mean cyclic displacements [u]cm and [w]cm in dependence of N
of cyclic CNL tests with rough plate, σn0 = 120 kPa, ηcm ≈ 12 ηpeak ; (left) on loose sand
(ID0 ≈ 30%) ; (right) on dense sand (ID0 ≈ 90%)
However, after excluding the first cycle of each package, the mean cyclic displacements
([u]cm and [w]cm ) in dependence of N on loose and dense sand were presented in Figure
2.93. The tests with steady amplitude (∆τ = 10 and 20 kPa in the identification series)
were also presented. In comparison with the steady amplitude of ∆τ = 10 and 20 kPa,
the evolution of [u]cm and [w]cm of the tests by changing the amplitude of cycles on both
densities were in the range of this boundary condition. In case of starting with amplitude
of cycle ∆τ = 20 kPa, the mean cyclic displacements [u]cm and [w]cm at N =10000 were
slightly larger than that with ∆τ = 10 kPa. Therefore, this can be attributed that the
change of amplitude of cycles between consecutive packages has no influence on the total
mean cyclic displacements [u]cm and [w]cm . Since this series was performed in the range of
ηcm < ηcar , if the significant change of amplitude of cycles, especially in case of ηcm close to
ηcar is applied, this would lead to different results (Wichtmann et al. [2010]).
139
0.8 1
monotonic ID0 | 30% ; Vn = 120 kPa ; ID0 | 30% ; Vn = 120 kPa ;
0.7 'W = 10 D 20 kPa ; CNL test 'W = 20 D 10 kPa ; CNL test
0.8
0.6 monotonic
K WVn
K WVn
0.5 0.6
post-cyclic phase ; 10-20LM120
0.4
0.3 0.4
0.2
0.2
0.1
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0 0
-0.1 monotonic
monotonic -0.2
-0.2
[u] (mm)
[u] (mm)

-0.3 -0.4
-0.4
-0.6
ID0 | 30% ; Vn = 120 kPa ;
-0.5 'W = 10 D 20 kPa ; CNL test ID0 | 30% ; Vn = 120 kPa ;
'W = 20 D 10 kPa ; CNL test
-0.6 -0.8
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.94: Post-cyclic phase of cyclic CNL tests by changing the amplitude of cycles on
loose sand (ID0 ≈ 30%) with rough plate, σn0 = 120 kPa, ηcm ≈ 21 ηpeak
After cyclic phase, the post-cyclic phase was then carried out. Figure 2.94 and 2.95 show
respectively the post-cyclic behaviour of interface on losse and dense sands. The values
of peak stress ratio (ηpeak ) at post-cyclic phase of these tests on both densities were not
different so much from those in monotonic tests. Most of them were slightly lower than those
in monotonic CNL tests. Altough during cyclic phase the interface behaved contractively
which induced more densification on loose sand (the occurence of slight dilation rate at the
beginning of post-cyclic phase) there was not the significant difference in ηpeak . The slight
decrease in ηpeak at post-cyclic phase on loose sand may be attributed to the high evolution
of [w]cm during cyclic phase (the magnitude of sliding displacement of interface, Uesugi
et al. [1989] and Fakharian & Evgin [1997]). Dense sand on which the interface behaved
contractively during cyclic loadig exhibited dilative behaviour at post-cyclic phase. The
dilation rate at post-cyclic phase was not so much different from monotonic test. The
slight decrease in ηpeak was also found at post-cyclic phase. This would be attributed to an
increase in crushing and wear of grains within the interface shear zone on dense sand.
140
1 1
ID0 | 90% ; Vn = 120 kPa ; ID0 | 90% ; Vn = 120 kPa ;
'W = 10 D 20 kPa ; CNL test 'W = 20 D 10 kPa ; CNL test
0.8 0.8
monotonic monotonic
K WVn
K WVn
0.6 0.6
post-cyclic phase ; 20-10DM120
post-cyclic phase ; 10-20DM120
0.4 0.4
0.2 0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.2 0.3
ID0 | 90% ; Vn = 120 kPa ;
0.2
0.1
monotonic
[u] (mm)
0.1
[u] (mm)
0 monotonic
0
-0.1 post-cyclic phase ; 20-10DM120
post-cyclic phase ; 10-20DM120 -0.1
-0.2
ID0 | 90% ; Vn = 120 kPa ; -0.2
-0.3 -0.3
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.95: Post-cyclic phase of cyclic CNL tests by changing the amplitude of cycles on
dense sand (ID0 ≈ 90%) with rough plate, σn0 = 120 kPa, ηcm ≈ 21 ηpeak
2.10 Cyclic interface direct shear tests on smooth plate
Cyclic interface direct shear tests on smooth plate were supplementarily performed in oder
to compare to those on rough plate. This series was mainly focused in CNS condition
(Table 2.6). In case of cyclic CNS condition on loose sand with σn0 = 60 kPa, there was
not any possible way to perform the tests whether with k = 1000 or 5000 kPa/mm. From
experimental observations in monotonic tests on smooth plate, the peak stress ratio ηpeak
on loose sand is very small. Obviously, under CNS condition the test with σn0 = 60 kPa,
these tests were only performed in a short period of time (τpeak ≈ 9 kPa at [w] ≈ 0.80
mm). Some of cyclic CNL tests were performed by observing some factors influencing on
the interface behaviour.
141
Table 2.6: Cyclic tests on smooth plate
N◦ ID0 σn k level of ηcm0 τ ηcm ∆τ
(%) (kPa) (kPa/mm) (-) (kPa) (-) (kPa)
10LL120 30 120 CNL 0 -5 < τ < 5 0 10

10LM120 30 120 CNL ≈ 1/2ηpeak 20 < τ < 30 0.21 10
10LH120 30 120 CNL ≈ 9/10ηpeak 40 < τ < 50 0.38 10
20LM120 30 120 CNL ≈ 1/2ηpeak 15 < τ < 35 0.21 20
10DL120 90 120 CNL 0 -5 < τ < 5 0 10
10DM120 90 120 CNL ≈ 1/2ηpeak 28 < τ < 38 0.28 10
10DH120 90 120 CNL ≈ 9/10ηpeak 50 < τ < 60 0.46 10
20DM120 90 120 CNL ≈ 1/2ηpeak 23 < τ < 43 0.28 20
10LL100 1000k 30 100 1000 0 5 < τ < 15 0.10 10
10LL100 5000k 30 100 5000 0 0 < τ < 10 0.05 10
10LM310 1000k 30 310 1000 ≈ 1/2ηpeak 42 < τ < 52 0.15 10
10LM310 1000k 30 310 1000 ≈ 2/3ηpeak 88 < τ < 98 0.30 10
10DM60 1000k 90 60 1000 ≈ 1/2ηpeak 10 < τ < 20 0.25 10
10DM100 1000k 90 100 1000 ≈ 1/2ηpeak 15 < τ < 25 0.20 10
10DM100 1000k 90 100 1000 ≈ 2/3ηpeak 30 < τ < 40 0.35 10
10DH100 5000k 90 100 5000 ≈ 2/3ηpeak 30 < τ < 40 0.35 10
20DH100 1000k 90 100 1000 ≈ 2/3ηpeak 25 < τ < 45 0.35 20
20DH100 5000k 90 100 5000 ≈ 2/3ηpeak 25 < τ < 45 0.35 20
10DH310 1000k 90 310 1000 ≈ 2/3ηpeak 105 < τ < 115 0.35 10
2.10.1 Cyclic CNL tests
Figure 2.96 shows the test results of cyclic interface tests under CNL condition on both
densities. As can be observed, the mean cyclic displacements ([u]cm and [w]cm ) obtained
from smooth plate were considerable than those from rough plate. Considering the normal
displacement [u]cm , unlike monotonic test it was found that cylic CNL tests on smooth
plate provided more contraction than rough plate did. The contractive behaviour was
found in the range of ηcm < ηcar . However, this was in contrast to those derived from
available investigations in which the normal displacement of rough plate was greater than
that of smooth plate (the effect of dilative behaviour which was mainly found on rough
plate influenced the volumetric behaviour and consequently resulted in more variation of
[u]).
142
0.0 0.1
ID0 | 90% ; Vn = 120 kPa ;
'W = 10 , 20 kPa ; CNL tests
-0.2
0.0
[u]cm (mm)
[u]cm (mm)
-0.4
ID0 | 30% ; Vn = 120 kPa ; -0.1
'W = 10 , 20 kPa ; CNL tests 10DL120
-0.6
10DM120
10LL120
10DH120
10LM120 -0.2 20DM120
-0.8 10LH120
20LM120
-1.0 -0.3
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
5 6
ID0 | 30% ; Vn = 120 kPa ; ID0 | 90% ; Vn = 120 kPa ;
'W = 10 , 20 kPa ; CNL tests 'W = 10 , 20 kPa ; CNL tests
5
4 10DL120
10LL120
10LM120 10DM120
[w]cm (mm)
4
[w]cm (mm)
10LH120 10DH120
3
20LM120 20DM120
3
2
2
1 1
0 0
1 10 100 1000 10000 1 10 100 1000 10000
N (-) N (-)
Figure 2.96: Evolution of mean cyclic displacements [u]cm and [w]cm in dependence of N
of cyclic CNL tests, σn0 = 120 kPa with smooth plate; (left) on loose sand (ID ≈ 30%) ;
(right) on dense sand (ID0 ≈ 90%).
For instance, Mortara [2001], Shahrour & Rezaie [1997] and Mortara et al. [2007] reported
that under cyclic CNL condition the normal displacement [u] from rough plate was larger
than that from smooth plate due to the effect of dilative behaviour. However, those obser-
vations were in displacement-controlled tests whereas this study was carried out by shear
stress-controlled tests. With displacement-controlled tests, a contracting behaviour followed
by dilation at each cycle was observed on rough plate (more evident on dense sand). Con-
sidering the shear stress-controlled tests, an evolution of accumulative displacements as a
function of N was obtained. In this study, it was found that the accumulation of dispace-
ments mainly depended on many factors such as initial normal stress (σn0 ), initial density
(ID0 ), mean cyclic stress ratio (ηcm ) and amplitude of cycle (∆τ ). When comparing two
different types of surface roughness, although the tests were performed at the same level of
ηcm (but ηcm of smooth plate was closer to ηpeak as a result of lower friction angle), the evo-
lution of mean cyclic shear displacement [w]cm on smooth plate was more significant than
that on rough plate and consequently led to more contraction. This displacement ([w]cm )
arose considerably, thought ηcm = 0 as shown in Figure 2.96. This can be deduced that
the smooth plate allowed the interface to slide than rough plate (the freedom in movement
143
of soil adjacent to the structure). The significant increase of mean cyclic shear displace-
ment [w]cm therefore induced more contractive behaviour (an increase in [u]cm ). Within
the range of ηcm < ηcar , the increase of cyclic amplitude (∆τ ) induced the increasing rate
of displacement intensity. In case of high level of ηcm on dense sand, the dilative behaviour
on smooth plate was less evident than that on rough plate.
At post-cyclic phase, the peak stress ratio (ηpeak ) was found to be slightly less than that
of monotonic test on both densities (Figure 2.97). It was also found that there was not
significant difference in peak stress ratio when performing the tests in different conditions.
Despite the increase in contraction during cyclic loading, the volumetric behaviour at post-
cyclic phase did not show any significant variation. Indeed, in the range of ηcm < ηcar cyclic
loading which induces more densification within the interface would at least lead to the
variation of volumetric behaviour at post-cyclic phase. This can be attributed to the effect
of dilative behaviour which can hardly be observed on smooth plate.
0.7 0.8
ID0 | 30% ; Vn = 120 kPa ; ID0 | 90% ; Vn = 120 kPa ;
0.6 'W = 10, 20 kPa ; CNL tests 'W = 10 , 20 kPa ; CNL tests
0.5 0.6
K WVn
K WVn
0.4
0.4
0.3
monotonic monotonic
0.2 10LL120 (cyclic phase) 10DL120 (cyclic phase)
0.2
0.1 10LM120 (cyclic phase) 10DM120 (cyclic phase)
20LM120 (cyclic phase) 20DM120 (cyclic phase)
0 10LL120 (post-cyclic phase) 0 10DL120 (post-cyclic phase)
-0.1
-0.2 -0.2
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.2 0.2
ID0 | 30% ; Vn = 120 kPa ; ID0 | 90% ; Vn = 120 kPa ;
0 'W = 10 , 20 kPa ; CNL tests monotonic 'W = 10 , 20 kPa ; CNL tests
-0.0
-0.2
monotonic post-cyclic phase ; 10DL120
[u] (mm)
[u] (mm)
post-cyclic phase ; 10LL120

-0.4
post-cyclic phase ; 10LM120 -0.2
-0.6
post-cyclic phase ; 10DM120
post-cyclic phase ; 20DM120
-0.8
-0.4
-1
post-cyclic phase ; 20LM120
-1.2 -0.6
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 2.97: Post-cyclic phase of cyclic CNL tests, σn0 = 120 kPa with smooth plate; (left)
on loose sand (ID0 ≈ 30%) ; (right) on dense sand (ID0 ≈ 90%).
144
Similarly to rough plate, the examination of grain breakage after testing was also carried out.
As can be well conceived that the smooth has less abrasive surface than rough plate and the
slippage generally occurs along the contact surface between smooth plate and sand (Uesugi
& Kishida [1986]). The tendency for grain crushing, due to the abrasion between grains and
smooth plate, within the localized shear zone would be hardly found. After cyclic loading
tests, it was found that smooth plate did not provide noticeable grain breakage. Figure
2.98 shows the grading curves of the tests performed on smooth plate in various conditions.
On loose sand the variation of grading curves representing the grain breakage could hardly
observed even considering the fine particle (D10 ) whereas slight variation of grading curves
could observed on dense sand.
100 100
90 90
80 80
finer by weight (%)
finer by weight (%)

70 70
60 60
50 50 ID0 = 90% ; Vn = 120 kPa ;
ID0 = 30% ; Vn = 120 kPa ;
smooth plate
smooth plate
40 40
30 30 initial (sieving method)
initial (sieving method)
monotonic CNL test
monotonic CNL test
20 20 10DL120
10LL120
10DM120
10 10LM120 10 20DM120
20LM120
0 0
0.01 0.1 1 10 0.01 0.1 1 10
(a) (b)
Figure 2.98: Grain size distribution of sand after testing with various CNL conditions on
smooth plate, σn = 120 kPa; (a) on loose sand (ID0 ≈ 30%); (b) on dense sand (ID0 ≈
90%)
2.10.2 Cyclic CNS tests
An important feature of cyclic interface tests is the CNS condition in which the degradation
of normal stress is described. Smooth plate also showed that the degradation of normal
stress was the main characteristic which was similar to that on rough plate. Having known
from experimental evidences, the significant degradation of normal stress was due to the
contraction within the interface. As can be seen on rough plate, there were many factors
that influenced the interface behaviour such as k, ID0 , σn0 , ηcm0 , ∆τ .
The results for cyclic CNS tests with smooth plate, σn0 = 100 kPa on dense sand are
typically presented as the diagrams of τ − σn , σn − [w] and [u] − [w] in Figure 2.99. These
results show the strong influence of k (k = 1000 and 5000 kPa/mm) on cyclic interface
behaviour. These can be compared with the tests given in Figure 2.77 for rough plate.
On dense sand, although these tests were performed at the same level of ηcm0 = 0.35, it
was evident that a number of cycles for reaching the critical state with smooth plate were
significantly less than those with rough plate. Considering the rate of degradation in normal
145
stress as a result of the progressive contraction, it was found that this rate of smooth plate
was slighly lower than that of rough plate. On rough plate, for instance in case of k = 1000
kPa/mm, σn dropped from 100 to 60 kPa during the first 100 cycles whereas σn dropped
from 100 to 80 kPa on smooth plate. However, the interesting feature of the degradation in
normal stress (and consequently the critical state was reached) was the level of initial mean
cyclic stress ratio (ηcm0 ).
While the critical state line of smooth plate is obviously lower than that of rough plate
(ηcrit = 0.43 for smooth plate whereas ηcrit = 0.566 for rough plate), with starting the same
level of ηcm , the stress state of smooth plate moves towards the critical state line due to the
degradation of normal stress and then early reaches the lower value of critical state line.
These two plates did not show the significant difference in volumetric behaviour ([u] − [w]
diagrams) especially in case of k = 5000 kPa/mm in which the normal displacement can be
deduced to be almost constant during cyclic loading.
To illustrate some of the main factors (i.e. σn0 , ηcm0 and ∆τ ) influencing the cyclic CNS
interface, first the tests performed with different values of initial normal stress (σn0 ) are
presented. Figure 2.100 shows the test results performed with σn0 = 60 kPa at ηcm0 =
0.25 and σn0 = 310 kPa at ηcm0 = 0.35 on dense sand. The influence of σn0 can well be
understood by considering the test results on rough plate as shown in Figure 2.80 and 2.82.
Obviously, the test performed with σn0 = 60 kPa provided less number of cycles for reaching
the critical state. A simimilar trend was also observed on smooth plate. With σn0 = 60
kPa, ηcm0 = 0.25 on smooth plate, a number of cycles for reaching the critical state were
N = 19 (in case of rough plate N = 1042 at ηcm0 = 0.35).
The effect of σn0 was more considerable when performing with σn0 = 310 kPa at the same
level of ηcm0 (e.g. ηcm = 0.35). As can be seen in Figure 2.82, the tendency to reach the
critical state on rough plate was hardly found (this test was performed until N = 32669
and then the post-cyclic phase was applied) whereas the smooth plate required 4,710 cycles
for reaching the critical state. The noticeable difference in number of cycles for reaching
the critical state of these two plates can be attributed to the difference in peak (ηpeak ) and
critical (ηcrit ) stress ratios. These two stress ratios obtained from smooth plate allowed the
stress state to reach the critical state early than rough plate did.
The effect of cyclic amplitude was also observed. Again, the effect of cyclic amplitude was
observed with ∆τ = 10 (30 < τ < 40) and 20 (25 < τ < 45) kPa, σn0 = 100 kPa, ηcm0 =
0.35, k = 1000 and 5000 kPa/mm. Figure 2.101 typically shows the test results in case of
∆τ = 20 (25 < τ < 45) kPa with k = 5000 kPa/mm. As can be seen on rough plate, since
the increase in amplitude of cycle was applied, a number of cycles for reaching the critical
state decreased obviously. In case of k = 5000 kPa/mm, ∆τ = 20 kPa, a number of cycles
N = 48 were required for reaching the critical state while N = 121 for ∆τ = 10 kPa. The
rate of degradation in normal stress was almost similar, the normal stress dropped from 100
to 80 kPa, but the stress state in case of ∆τ = 20 kPa could reach the critical state early
because of the size of cycles (see τ − σn planes).
146
100 100
N = 1-100 ID0 | 90 % ; Vn0 = 100 kPa ; Kcm0 = 0.35 N = 1-50 ID0 | 90 % ; Vn0 = 100 kPa ; Kcm0 = 0.35
N = 150-160 k = 1000 kPa/mm ; 30< W<40 kPa N = 90 - 100 k = 5000 kPa/mm ; 30< W<40 kPa
80 N = 200-210 80 N = 110-121
N = 245-248
60 60
W (kPa)
W (kPa)
1 1
40 40
20 Kcrit = 0.43 20
1 Kcrit = 0.43
1
0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
Vn (kPa) Vn (kPa)
140 140
ID0 | 90 % ; Vn0 = 100 kPa ; Kcm0 = 0.35 ID0 | 90 % ; Vn0 = 100 kPa ; Kcm0 = 0.35
k = 1000 kPa/mm ; 30< W<40 kPa k = 5000 kPa/mm ; 30< W<40 kPa
120 120
N = 1-50
100 N = 90 - 100
100
N = 110-121
Vn (kPa)

Vn (kPa)
80 80
60 60
N = 1-100
N = 150-160
40 N = 200-210 40
N = 245-247
20 towards critical state 20
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w] (mm) [w] (mm)
0.02 0.04
ID0 | 90 % ; Vn0 = 100 kPa ; Kcm0 = 0.35
ID0 | 90 % ; Vn0 = 100 kPa ; Kcm0 = 0.35
k = 1000 kPa/mm ; 30< W<40 kPa
0 k = 5000 kPa/mm ; 30< W<40 kPa
0.02
[u] (mm)
-0.02
[u] (mm)
0
-0.04
-0.02
-0.06 N = 1-100 N = 1-50
N = 150-160 N = 90 - 100
N = 200-210 -0.04
-0.08 N = 110-121
N = 245-247
-0.1 -0.06
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w] (mm) [w] (mm)
Figure 2.99: Influence of k on cyclic CNS tests, σn0 = 100 kPa, ηcm0 = 0.35 on dense sand
(ID0 ≈ 90%) with smooth plate; (left) k = 1000 kPa/mm; (right) k = 5000 kPa/mm
147
80 300
N = 1-100
k = 1000 kPa/mm ; 10< W<20 kPa k = 1000 kPa/mm ; 105< W<115 kPa
N = 190-200
N = 500-510
60 N = 1-10 N = 1000-1010
N = 11-15 N = 2000-2010
200
N = 16-19 N = 3000-3010
W (kPa)
N = 4000-4010
W (kPa)

N = 4760-4710
40 Kpeak = 0.49
Kpeak = 0.49 towards critical state
1 1
Kcrit = 0.43 100
1
20
Kcrit = 0.43
1
0 0
0 20 40 60 80 100 0 100 200 300 400
Vn (kPa) Vn (kPa)
100 350
k = 1000 kPa/mm ; 10< W<20 kPa 300 k = 1000 kPa/mm ; 105< W<115 kPa
80
N = 1-10
N = 11-15 250
N = 16-19
Vn (kPa)
200
Vn (kPa)
N = 1-100
N = 190-200
150 N = 500-510
40 N = 1000-1010
N = 2000-2010
100
N = 3000-3010
20 N = 4000-4010
50 N = 4760-4710
0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12 14 16
[w] (mm) [w] (mm)
0.02 0.04
ID0 | 90 % ; Vn0 = 60 kPa ; Kcm0 = 0.25 N = 1-100
k = 1000 kPa/mm ; 10< W<20 kPa 0.02 N = 190-200
0 N = 500-510
N = 1000-1010
0 N = 2000-2010
-0.02 N = 3000-3010
-0.02 N = 4000-4010
[u] (mm)
[u] (mm)
N = 4760-4710
-0.04 -0.04 towards critical state
-0.06
-0.06
N = 1-10
-0.08
N = 11-15
-0.08 N = 16-19 ID0 | 90 % ; Vn0 = 310 kPa ; Kcm0 = 0.35
-0.1
towards critical state k = 1000 kPa/mm ; 105< W<115 kPa
-0.1 -0.12
0 2 4 6 8 10 12 0 2 4 6 8 10 12 14 16
[w] (mm) [w] (mm)
Figure 2.100: Influence of σn0 on cyclic CNS tests, k = 1000 kPa/mm, on dense sand (ID0 ≈
90%) with smooth plate; (left) σn0 = 60 kPa at ηcm0 = 0.25; (right) σn0 = 310 kPa at ηcm0
= 0.35
148
100 140
ID0 | 90 % ; Vn0 = 100 kPa ; Kcm0 = 0.35
N = 1-20 ID0 | 90 % ; Vn0 = 100 kPa ; Kcm0 = 0.35
k = 5000 kPa/mm ; 25< W<45 kPa
k = 5000 kPa/mm ; 25< W<45 kPa 120
N = 25-35
80
N = 40-48
100
60
W (kPa)
Vn (kPa)
Kpeak = 0.49 80
1
40 60
N = 1-20
N = 25-35
40
N = 40-48
Kcrit = 0.43 20
1
0 0
0 20 40 60 80 100 120 140 0 2 4 6 8 10 12 14
Vn (kPa) [w] (mm)
(a) (b)
Figure 2.101: Cyclic CNS test on smooth plate, σn = 100 kPa, ∆τ = 20 kPa, k = 1000
kPa/mm on dense sand (ID0 ≈ 90%); (a) τ − σn plane; (b) σn − [w] plane
When considering the evolution of ηcm during cyclic phase, Figure 2.102 illustrates how the
stress state represented by ηcm , corresponding to the degradation of normal stress, moves
towards the critical state line. The evolution of ηcm corresponding to the degradation of σcm
as a function of N as well as [w]cm clearly explains how the critical state of the interface
is reached under various conditions. Similarly to cyclic CNS tests on rough plate, the
degradation in normal stress increased as a function of N and [w]cm . When comparing this
degradation between the tests performed at different levels of ηcm (e.g. the tests with σn0 =
100 kPa at ηcm0 = 0.20 and 0.35), the rate of normal stress deradation at ηcm0 = 0.35 was
lower than that at ηcm0 = 0.20 but the stress state at ηcm0 = 0.35 could reach the critical
state with N = 248 whereas in case of ηcm0 = 0.20 required 294 cycles. The test performed
with low initial stress ratio (ηcm0 ) required a large number of cycles for reaching the critical
state line.
On loose sand, smooth plate provided the significant degradation of σn as a result of the
progressive contraction. Figure 2.103 shows typical results due to the effect of imposed
normal stiffness on loose sand with σn0 = 100 kPa, ∆τ = 10 kPa. These two tests were
performed at almost the same level of ηcm0 (k = 1000 kPa/mm at ηcm0 = 0.10, 5 < τ <
15 and k = 5000 kPa/mm at ηcm0 = 0.05, 0 < τ < 10). Although the levels of ηcm0 of
these two tests were very small, the stress state could evolve only 3 cycles for k = 1000
kPa/mm whereas in case of k = 5000 kPa/mm the stress state could evolve only one cycle.
The significant degradation of normal stress in these two tests dropped rapidly during the
first cycle (especially in case of k = 5000 kPa/mm, σn dropped from 100 to 40 kPa) and
consequently the stress state moved to the critical state line.
149
500 1
ID0 | 90% ; k = 1000 kPa/mm ; smooth plate D0||90%
IID0 90%;;kk==1000
1000kPa/mm
kPa/mm;;smooth
smoothplate
plate
Vn0 = 60 kPa
10sn60_1000k ; 'W = 10 kPa ; Kcm0 = 0.25 VVn0n0==60 kPa;;'W
60 kPa
10sn60_1000k 'W==10 kPa;;KKcm0
10kPa cm0==0.25
0.25
Vn0 = 100 kPa ; 'W = 10 kPa ; Kcm0 = 0.20
10sn100_1000k VVn0n0==100 kPa;;'W
100kPa
10sn100_1000k 'W==10 kPa;;KKcm0
10kPa cm0==0.20
0.20
400 V = 100 kPa ; 'W = 20 kPa ; Kcm0 = 0.35 0.8 VVn0n0==100 kPa;;'W
100kPa 'W==20 kPa;;KKcm0
20kPa
20sn100_1000k
n0 20sn100_1000k cm0==0.35
0.35
Vn0 = 100 kPa ; 'W = 10 kPa ; Kcm0 = 0.35
10sn100_1000k VVn0n0==100
10sn100_1000kkPa;;'W
100kPa 'W==10 kPa;;KKcm0
10kPa ==0.35
0.35
Kcm WcmVn cm (-)

cm0
Vn0 = 310 kPa ; 'W = 10 kPa ; Kcm0 = 0.35
10sn310_1000k VVn0n0==310 kPa;;'W
310kPa
10sn310_1000k 'W==10 kPa;;KKcm0
10kPa cm0==0.35
0.35
Vn cm (kPa)
300 0.6
Kpeak = 0.49
Kcrit = 0.43
200 0.4
100 0.2
0 0
0 1 10 100 1000 10000 0 1 10 100 1000 10000
N (-) N (-)
500 1
ID0 | 90% ; k = 1000 kPa/mm ; smooth plate ID0 | 90% ; k = 1000 kPa/mm ; smooth plate
Vn0 = 60 kPa
10sn60_1000k ; 'W = 10 kPa ; Kcm0 = 0.25 Vn0 = 60 kPa ; 'W = 10 kPa ; Kcm0 = 0.25
10sn60_1000k
Vn0 = 100 kPa ; 'W = 10 kPa ; Kcm0 = 0.20
10sn100_1000k Vn0 = 100 kPa ; 'W = 10 kPa ; Kcm0 = 0.20
10sn100_1000k
400 Vn0 = 100 kPa ; 'W = 20 kPa ; Kcm0 = 0.35
20sn100_1000k
0.8 Vn0 = 100 kPa ; 'W = 20 kPa ; Kcm0 = 0.35
10sn100_1000k
Vn0 = 100 kPa ; 'W = 10 kPa ; Kcm0 = 0.35
10sn100_1000k Vn0 = 100 kPa ; 'W = 10 kPa ; Kcm0 = 0.35
10sn100_1000k
V n0 = 310 kPa ; 'W = 10 kPa ; Kcm0 = 0.35 V n0 = 310 kPa ; 'W = 10 kPa ; Kcm0 = 0.35
Kcm WcmVn cm (-)
10sn310_1000k 10sn310_1000k
300
Vn cm (kPa)
0.6
Kpeak = 0.49
200 0.4 Kcrit = 0.43
100 0.2
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
[w]cm (mm) [w]cm (mm)
Figure 2.102: Degradation of σn cm corresponding to the evolution of ηcm of cyclic CNS

tests on dense sand (ID0 ≈ 90%) with smooth plate
The influence of ηcm0 with the same value of σn0 on loose sand was also observed by
performing cyclic CNS tests with σn0 = 310 kPa, k = 1000 kPa/mm, the first one was
performed at ηcm0 = 0.30 as described on rough plate (see Figure 2.76). This was evident
in number of cycles for reaching the critical state (see Figure 2.104). On smooth plate, the
normal stress decreased rapidly during the evolution of the first cycle (σn dropped to 230
kPa) and then the stress state could not evolve the next cycle whereas on rough plate the
normal stress decreased continuously until σn = 165 kPa at N = 2035. This can clearly
be described by the peak and critical stress ratios (ηpeak and ηcrit ) of smooth plate which
are lower than those of rough plate. On smooth plate, since the stress state was closed to
the critical state line at the beginning of cyclic loading phase, it could move quickly to the
critical state line during the first cycle. However, when performing the test at ηcm0 = 0.15,
σn0 = 310 kPa on smooth plate it was found that the normal stress decreased continously
as a function of N and [w] until σn = 120 kPa at N = 481.
150
100 100
80 N = 1-3 80
N=1
60 60
W(kPa)
W (kPa)
40 40 Kpeak = 0.43
Kpeak = 0.43 1
1
20 20
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Vn (kPa) Vn (kPa)
140 140
120 120
N = 1-3 N=1
100 100
80 80
Vn (kPa)
Vn (kPa)
60 60
40 40
20 20
0 0
0 1 2 3 4 5 0 0.4 0.8 1.2 1.6 2
[w] (mm) [w] (mm)
0 0.02
0
N=1
-0.05
-0.02
[u] (mm)

[u] (mm)
-0.1 -0.04
-0.06
-0.15
-0.08
-0.2 -0.1
0 0.5 1 1.5 2 2.5 3 0 0.4 0.8 1.2 1.6 2
[w] (mm) [w] (mm)
Figure 2.103: Influence of k on cyclic CNS tests, σn0 = 100 kPa, loose sand (ID0 ≈ 30%)
with smooth plate; (left) k = 1000 kPa/mm at ηcm0 = 0.10 ; (right) k = 5000 kPa/mm at
ηcm0 = 0.05
151
300 300
N=1 N = 1-50
towards critical state N = 100-110
200 200 N = 190-200
N = 300-310
N = 400-410
W (kPa)
W (kPa)
N = 470-481
towards critical state Kpeak = 0.43
Kpeak = 0.43
1
1
100 100
0 0
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
400 400
N = 1-50
300 300
N = 100-110
N=1
N = 190-200
N = 300-310
Vn (kPa)
Vn (kPa)
N = 400-410
200 200
N = 470-481
100 100
0 0
0 2 4 6 8 10 0 2 4 6 8 10 12
[w] (mm) [w] (mm)
0.04 0.05
0
N=1 N = 1-50
towards critical state -0.05 N = 100-110
-0.04 N = 190-200
[u] (mm)
[u] (mm)
N = 300-310
-0.1
N = 400-410
-0.08
N = 470-481
-0.15 towards critical state
-0.12
-0.2
-0.16 -0.25
-0.2 -0.3
0 2 4 6 8 10 12 0 2 4 6 8 10 12
[w] (mm) [w] (mm)
Figure 2.104: Influence of ηcm0 on cyclic CNS tests, σn0 = 310 kPa, loose sand (ID0 ≈ 30%)
with smooth plate; (left) at ηcm0 = 0.30 ; (right) at ηcm0 = 0.15
152
500 500
ID0 | 30% ; smooth plate ID0 | 30% ; smooth plate
Vn0 = 100 kPa; k = 1000 kPa/mm; 'W = 10 kPa ; Kcm0 = 0.10
10sn100_1000k Vn0 = 100 kPa; k = 1000 kPa/mm; 'W = 10 kPa ; Kcm0 = 0.10
10sn100_1000k
10sn100_1000k
400 Vn0 = 310 kPa; k = 1000 kPa/mm; 'W = 10 kPa ; Kcm0 = 0.15
10sn310_1000k
400 Vn0 = 310 kPa; k = 1000 kPa/mm; 'W = 10 kPa ; Kcm0 = 0.15
10sn310_1000k
10sn310_1000k
Vn cm (kPa)
Vn cm (kPa)
300 300
200 200
100 100
0 0
0 1 10 100 1000 10000 0.0 0.5 1.0 1.5 2.0 2.5 3.0
N (-) [w]cm (mm)
0.8 0.8
ID0 | 30% ; smooth plate ID0 | 30% ; smooth plate
10sn100_1000k
10sn100_1000k
10sn310_1000k
0.6 0.6
Vn0 = 310 kPa; k = 1000 kPa/mm; 'W = 10 kPa ; Kcm0 = 0.30 Vn0 = 310 kPa; k = 1000 kPa/mm; 'W = 10 kPa ; Kcm0 = 0.30
Kcm WcmVn cm (-)
Kcm WcmVn cm (-)
10sn310_1000k 10sn310_1000k

0.4 Kcrit = 0.43
0.4 Kcrit = 0.43
0.2 0.2
0 0
0 1 10 100 1000 10000 0.0 0.5 1.0 1.5 2.0 2.5 3.0
N (-) [w]cm (mm)
Figure 2.105: Degradation of σn cm corresponding to the evolution of ηcm of cyclic CNS

tests on loose sand (ID0 ≈ 30%) with smooth plate
Again, the degradation of σn on loose sand corresponding to the evolution of ηcm0 as shown
in Figure 2.105 can clarify the effect of k in various conditions. Since the rate of degradation
in normal stress increased as a function of N and [w] and then the stress state represented by
ηcm moved towards critical state line. From experimental observations, it can be deduced
that the worse condition of cyclic CNS interface shear tests is the test performing with
smooth plate, low initial normal stress, high level of ηcm0 and high value of k.
153
2.11 Conclusions
Experimental campaigns carried out in this study mainly focus on cyclic direct shear tests
on sand (Fontainebleau)/rough material interface under constant normal load (CNL) and
constant normal stiffness (CNS) conditions. These tests were intended to simulate the
situation along the shaft of piles subjected to a large number of cycles due to environmental
or anthropic loadings. The rough plates which were mainly used in this study were made
by gluing on a steel plate of a mixture of araldite (epoxy) and Fontainebleau sand. In case
of smooth plate, in comparison with rough plate, some tests were additionally performed.
Monotonic tests preliminarily performed in order to determine the peak (δpeak ), critical
(δcrit ) and characteristic (δcar , separating contractive and dilative domains) friction angle
allowed us to decide the cyclic test campaigns. Three levels of mean cylic stress ratio (ηcm ,
e.g. ηcm = 0, ≈ 21 ηpeak and ≈ 9/10ηpeak ) were then performed.
The cyclic tests (the shear stress-controlled tests) were classified into two groups for two
purposes:
• Identification, CNL (constant normal stress) tests
• Validation of constitutive law, CNS (constant normal stiffness) tests
All tests were planned to 104 cycles, unless early termination occurred (this phenomenon
was found in case of ηcm close to ηpeak ).
In particular, a special care in analysing the leakage of material during the test was then
devoted in order to get the proper results. In a reasonable manner, a simple inclusive correc-
tion was then be performed by considering the leakage of sand as the fictitious contraction
proportional to the accumulated [w].
The cyclic tests performed with various conditions provided the deep insight of interface
behaviour. From experimental observations in this study, the main factors playing an
important role in cyclic interface shear tests, whether under CNL or CNS conditions, can
be summarized as follows:
• Cyclic CNL condition
With rough plate, it was found that the interface behaved contractively within the range
of ηcm < ηcar whether on loose or dense sands. The intensity of mean cyclic displacements
([u]cm and [w]cm ) increased as a function of ηcm and ∆τ (when performing the tests at ηcm
close to ηcar , this would lead to different results). The volumetric behaviour of cyclic test
where 104 cycles were reached was more contractive than monotonic test. The rate of the
intensity of [u]cm and [w]cm decreased as a function of N as well as ID0 . The tests performed
with high value of σn provided less intensity of [u]cm and [w]cm because the interface was
restrained to expand when applying the high value of σn . In case of ηcm close to ηpeak ,
dense sand showed the existence of characteristic state (passage from contractive to dilative
behaviour) proposed by Luong [1980]. A number of cycles for reaching the critical state (the
state that the cycles could not evolve anymore) decreased when increasing the amplitude
of cycles (∆τ ).
154
The post-cyclic phase under CNL condition showed slight variation of peak stress ratio
(ηpeak ) in comparison with that from monotonic test. When performing with low initial
normal stress, ηpeak at post-cyclic phase was slightly higher than that from monotonic test
while high initial normal stress provided slighly lower ηpeak . Considering the grain breakage
within the interface shear zone represented by the grain size distribution curves (sieving
method), it was found that the fine particle size (D10 ) would be appropriate to represent
the change in grain breakage analysis. Within the range of ηcm < ηcar , an increase in σn ,
ηcm and ID0 provided more grain breakage (a shift of grading curves to the left).
In case of cyclic CNL tests by changing the sequence and amplitude of two consecutive
packages (∆τ = 10 and 20 kPa), it was found that the change of amplitude of cycles
between consecutive packages had no influence on the total mean cyclic displacements [u]cm
and [w]cm . However, if the significant change of amplitude of cycles, especially in case of
ηcm close to ηcar is applied, this would lead to different results.
With smooth plate, there existed the similar trend as can be found on rough plate. The
intensity of [u]cm and [w]cm still increased as a function of ηcm and ∆τ within the range of
ηcm < ηcar on both densities. However, the intensity of the displacement with smooth plate
was found to be more considerable than that with rough plate. This was in contrast to the
available investigations in which the normal displacement of rough plate was greater than
that of smooth plate due to the influence of dilative behaviour which was mainly found on
rough plate. In comparison between two different types of surface roughness, smooth plate
which provided lower friction angle allowed the mean cyclic shear displacement ([w]cm ) to
evolve considerably (due to ηcm close to ηpeak ). Therefore, the significant increase of [w]cm
induced more contractive behaviour (an increase in [u]cm ). In case of high level of ηcm on
dense sand, the dilative behaviour on smooth plate was less evident than that on rough
plate.
Smooth plate showed slight decrease in peak stress ratio (ηpeak ) at post-cyclic phase on
both densities. Altough during cyclic phase the interface behaved contractively and conse-
quently the interface exhibited more densification, there was not any significant variation in
volumetric behaviour at post-cyclic phase. The examination of grain breakage after testing
was found that the variation of grading curves could hardly observed on loose sand even
considering the fine particle (D10 ) whereas slight variation of grading curves could observed
on dense sand. This is due to less abrasive surface of smooth plate. Generally, the slippage
occurs along the contact surface between smooth plate and sand instead of an intense shear
zone.
• Cyclic CNS condition
The great interest of the loading paths with an imposed stiffness (k) is the variation of
normal stress acting on the interface. This variation of the normal stress mainly depends on
the volumetric behaviour of the granular soil adjacent to the structure (either to contract or
to dilate). Under cyclic CNS loading, the significant degradation in normal stress associated
with the degradation in shear stress was observed as a result of contraction during shearing.
With rough plate, the significant degradation of normal stress as a result of the progressive
contraction accompanied with N and [w] led to the critical state easily. From experimental
155
observations in this study, an increase of normal stiffness (k), cyclic amplitude (∆τ ) and
level of mean cyclic stress ratio (ηcm ) induced the gradual degradation in normal stress and
consequently the stress state moved towards the critical state. The interface could reach
the critical state easily (with a few number of cycles) on loose sand whereas dense sand
which exhibited low attitude to contract required a large number of cycles for reaching the
critical state.
During cyclic shearing, the degradation of σn as a function of N and [w] induced an increase
in ηcm which evolved and consequently moved towards the critical state line.
Smooth plate which obviously showed lower value in peak friction angle (the effect of dilative
behaviour) provided less number of cycles for reaching the critical state. The significant
difference in number of cycles for reaching the critical state of two different plates was found
when performing the tests with the same condition. The worse condition which the interface
could reach the critical state easily was that the test was performed on loose sand with low
initial normal stress (σn0 ), high value in normal stiffness (k), high level in mean cyclic stress
ratio (ηcm ) and large-amplitude cycles (∆τ ).
Considering the grain breakage within the interface after testing, because most of the tests
in cyclic CNS condition could be performed in a short period the grain breakage represented
by the change in fine particle size (D10 ) could hardly be observed (especially on loose sand).
However, when performing the test with rough plate, high initial normal stress (σn0 ), low
value in normal stiffness (k) and low level in mean cyclic stress ratio (ηcm ) on dense sand,
the test could continuously be performed and then provided the grain breakage (the shift
of grading curves to the left).
156
Chapter 3
Constitutive modeling of direct

shear tests
3.1 Introduction
The constitutive modeling of element tests generally refers to elasto-plasticity, elasto-visco-

plasticity, rate-type approach, defined by relations between time derivation of loading and
time derivative of response and related evolution of state or memory parameters. These
relations are integrated by the increments of time in order to build any finite paths, the
measurable variables being stresses, strains and relative displacements, i.e. Desai et al.
[1985] ; Aubry et al. [1990] ; Shahrour & Rezaie [1997] ; Mortara [2001] ; Gennaro & Frank
[2002] ; Mortara et al. [2002] ; Boulon et al. [2003] ; D’Aguiar et al. [2008] ; D’Aguiar et al.
[2011] including the others, a generalized plasticity (Liu et al. [2006] ; Liu & Ling [2008]).
The traditional approach which has been developed in Lab 3S-R is the rate-type (after
called incremental) analysis proposed by Boulon [1989].
In this study, a rate-type approach limited to some special paths in case of soil-structure
interaction is developed: the direct shear paths at constant normal stress (CNL) and at
prescribed normal stiffness (CNS, prescribed normal stiffness k) are previleged. It is noticed
that the CNL paths are in fact CNS paths with k = 0.
Considering only special paths, Boulon [1988] tried to model no more infinitesimal incre-
ments, but if possible finite increments, as large as possible in order to make possible finite
element computations with a limited volume and time of calculation.
In case of monotonic CNL and CNS tests, it has been possible to directly model analytically
entire path. Each path is defined by 3 parameters: the initial density (ID0 ), the initial
normal stress (σn0 ) and the prescribed normal stiffness (k). In addition, the length of the
path in terms of tangential displacement ([w]max ) is also considered.
In case of cyclic CNL and CNS tests, the situation is much more complex, due to the number
of parameters to be taken into account (i.e. initial density (ID0 ), the initial mean cyclic
157
normal stress (σn cm0 ), the initial mean cyclic shear stress (τcm0 ), the amplitude of cyclic
shear stress (∆τ ), and the number of cycle N ). In this study, the initial mean cyclic stress
ratio (ηcm0 = στncm0
cm0
) instead of τcm0 , and the amplitude of cyclic stress ratio (∆η = σn∆τ
cm0
)
would be preferable.
It has been possible to model analytically the CNL tests while the CNS tests have been
modeled by finite analytical increments according to the following principle. One CNS finite
increment is a summation of a CNL finite increment and of a finite increment in compression
or extension (Figure 3.1).
Vn cm
CNL
compression
(or extension)
[w]cm
Figure 3.1: Derivation of CNS path from a CNL finite increment
In addition, the memory variable allowing to chain the finite increments is the mean cyclic
normal relative displacement [ucm ] which is equivalent to the interface weight density under
normal stress (γd σn ).
In this chapter, one of the principal categories of constitutive modeling is detailed. The
incremental (rate type) model is first introduced and discussed. This model serves as the
implicit approach for describing the monotonic interface behaviour. After that, a study on
modeling the visco-plastic behaviour of the interface subjected to a large number of cycles
is proposed. The so-called explicit model (Wichtmann [2005], Wichtmann et al. [2010],
Niemunis et al. [2005]) in which the number of cycles N is treated instead of time is then
detailed. However, the strategy of model calibration is only detailed and provides a first step
to check the consistency of the model on rough plates. Simulations of interface behaviour
are then compared with experimental results and discussed.
3.2 Monotonic tests
3.2.1 Incremental (rate type) constitutive models
A constitutive relation within the framework of non-linear incremental (or rate type) law
has successively been carried out at laboratory 3S-R (e.g. Boulon [1989] ; Hoteit [1990] ;
158
Boutrif [1993] ; Garnica-Anguas [1993]). In this frame work, unlike elasto-plastic law, the
interface is considered as being a surface of kinematic discontinuity. The interface thickness
is not taken into account as a constitutive parameter. In general, the interface thickness
is assumed to be proportional to the value of D50 . In other words, it would be said that
the interface is considered as a zero thickness within the non-linear incremental framework
(Boulon & Nova [1990], Boulon et al. [1995]).
Having mentioned in the first chapter (Constitutive models section), the incremental inter-
face model is capable of describing a general direct shear paths which can govern the set of
possible responses of the interface in case of two-dimensional condition.
By definition, a constitutive law is a functional relation which allows to express the internal
forces (stress vector for the interface, [σ]), depending on the kinematic discontinuities at
the contact (relative displacement vector to the interface, [u]) that describe the complex
deformations of the medium;

[w] τ
[u] = and [σ] =
[u] σn
Again, the basic feature of this constitutive equation based on the non-linear relation be-
tween the incremental loadings and the corresponding incremental response shown in (1.33)
is expressed by:

τ̇ τ̇ ([ẇ] , [u̇] , memory parameters)
=
σ̇ σ̇([ẇ] , [u̇] , memory parameters)
Several kinds of interpolation fucntion (e.g. polynomial, trigonometric, exponential func-

tions etc. proposed by Boulon [1989] ; Boutrif [1993] ; Garnica-Anguas [1993]) are then
described for constructing an approximate incremental response.
Finally, the constitutive equation can formally be written via Euler’s theorem for homoge-
neous functions (1.42):
" #
∂ τ̇ ∂ τ̇
τ̇ ∂[ẇ] ∂[u̇] [ẇ]
= ∂ σ˙n ∂ σ˙n
σ˙n ∂[ẇ] ∂[u̇]
[u̇]
3.2.2 Analytical approach
• Key parameters
In this approach, the basic key parameters which are obtained by the derivation of interface
direct shear paths (representing η = τ /σn and [u] as a function of [w] for CNL condition; τ
and σn as a function of [w] for monotonic CNS condition) are presented by Figure 3.2 and
3.3, respectively.
159
W [u ] [u]crit
W peak Z
d [u ]
d [ w]
W crit [u] peak

>w@
[ w] peak [ w]crit
dW
D0
d [w]0 >w@ Z0
d [u ]
d [ w] 0
[ w] peak [ w]crit
(a) (b)
Figure 3.2: Key parameters for description of CNL test
To highlight the physical phenomena commonly encountered during shear loading (the
dilatancy, the evolution of friction angle, the grain breakage, etc.), these two figures show
the dependence of the evolution of the interface behaviour. They express the existence of
two phases of concurrent behaviour and seem difficult to separate:
1. The first phase is the phase before peak stress ratio (ηpeak = τpeak /σn ) where the
evolution of the tangential relative displacement ([w]) is not dominant.
2. The second phase corresponds to the post-peak (ηpeak ) and the interface behaviour
evolves as a function of the tangential relative displacement ([w]) for both conditions.
W Vn
W peak
dW V n peak
D
d [w] dV n
E
d [w]
Vn0
dV n
E0
d [w] 0
dW
D0
d [w]0 >w@ >w@
[ w] peak [ w] peak
(a) (b)
Figure 3.3: Key parameters for description of CNS test
• at constant normal stress (CNL), the presence of the critical state of shear stress
(τcrit ) preceeded by softening behaviour and a continuous evolution of the normal
displacement ([u]) at large tangential displacement ([w]) can be observed.
• at an imposed normal stiffness (CNS), τ and σn tend to decrease or increase and
160
the interface exhibits either a contractive or dilative behaviour (depending on
the initial density) with the reduction rate as a function of tangential relative
displacement.
In general, there exists a stabilisation of the normal displacement ([u], without dilation or
contraction) at large tangential displacement ([w]) in accordance with a critical state con-
cept. However, from typical interface test results, it is found that the feature of [u]-[w] and
σn − [w] diagrams are almost the same under CNS condition. The principal characteristics
of the interface are the initial contraction or dilation (depending on the initial density) due
to the rearrangement of the particles of the sand and the decrease after the peak. The last
one would probably be due to either a degradation related to the grains breakage within
the interface or the loss of material during the test (Boulon [1988]).
Therefore, during shear loading, the interface behaviour would be considered in two parts:
• The first part, due to rearrangement of the particles of the sand within the interface
shear zone during shear loading (contraction-dilation), is only a function of shear
relative displacement.
• The second part is due to the grain breakage linked with the specific interface energy,
W . After the peak, normal stress allows a linear slope (β+ ) with respect to W .
However, in a certain area of W , the slope β+ = ∂σn /∂[w]+ seems constant for an initial
normal stress (σn0 ) and initial density (ID0 ). In other words, the energy as a result of shear
loading is not high enough to allow the grain breakage. The influence of grain breakage on
the path of σn can be illustrated by Figure 3.4.
Vn
without grain breakage
with grain breakage
>w@
Figure 3.4: Interface direct shear test under CNS condition
With the hypothesis that the grains within the localized shear zone has a tendency to
be crushed during shearing load, and then the progress of crushing grains leads to more
contraction. Garnica-Anguas [1993] proposed the originality of the formulation describing
the grain breakage within the interface. This formulation can be expressed as:
σr = β+ (σn0 , ID0 ) W (3.1)
161
where:
σr is the normal stress due to the grain breakage.
β+ is the slope (∂σn /∂[w]+ ) at large tangential displacement.
W is the specific energy.
Figure 3.5 illustrates some of typical key parameters chosen to describe the interface be-
haviour under both conditions. From data of a typical interface direct shear tests shown
in [u] − [w] diagram (see also [u] − [w] diagram in the previous chapter), it is also worth
mentioning that the presence of the slight evolution of normal displacement ([u]) at large
tangential displacement ([w]) is observed. This may lead to a controversy with that of crit-
ical state concept in which the normal displacement is almost constant. However, with the
concept of grain breakage, this can be satisfactory for describing the interface behaviour.
When first considering the direct shear path under CNL condition, basic key parameters
are proposed to describe these curves;
• Before-peak phase:
– the initial modulus of η([w]), α0 : α0 = ∂η/∂[w]0
– peak stress ratio, ηpeak
– shear relative displacement corresponding to ηpeak , [w]η peak
– the initial dilatancy, ω0 : ω0 = ∂[u]/∂[w]0
• Post-peak phase:
– critical stress ratio, ηcrit
– the normal relative displacement at critical state, [u]crit
– the dilatancy, ω+ , corresponding to the post-peak phase: ω+ = ∂[u]/∂[w]+
Similarly, under CNS condition the basic key parameters describing the evolution of shear
(τ ) and normal σn stresses are then defined:
• Before-peak phase:
– the initial modulus of η([w]), α0 : α0 = ∂η/∂[w]0
– peak stress ratio, ηpeak
– peak normal stress, σn peak
– shear relative displacement corresponding to ηpeak and σn peak , [w]η peak and
[w]σ peak
– the initial modulus of σn ([w]), β0 : β0 = ∂σn /∂[w]0
– the initial dilatancy, ω0 : ω0 = ∂[u]/∂[w]0
162
1
ID0 | 90 % ; CNL ; rough plate
K peak Vn0 = 60 kPa
0.8
K crit
K WVn
0.6
0.4
0.2
wK
D0
w[w]0 [ w]crit 3[ w] peak
0
0 [ w] peak 1 2 3 4 5 6
[w] (mm)
0.25
Vn0 = 60 kPa ; k = 1000 kPa/mm
[u]crit w[u ]
0.15 Z
w[ w]
[u] (mm)
0.05
[u] peak
w[u ]
-0.05 Z0
w[ w]0
[ w]crit 3[ w] peak
-0.15
0 [ w] peak 1 2 3 4 5 6
[w] (mm)
250
Vn = 60 kPa ; k = 1000 kPa/mm
V n peak
200 wV n
E
w[w]
Vn (kPa)
150
100
50 wV n
E0
w[w]0
[ w] peak
0
0 1 2 3 4 5 6
[w] (mm)
Figure 3.5: Illustration of typical key parameters on dense sand (ID0 ≈ 90%), σn = 60 kPa,
CNL and CNS (k = 1000 kPa/mm) with rough plate
163
• Post-peak phase:
– critical stress ratio, ηcrit
– the modulus of σn ([w]), β+ , corresponding to the post-peak phase : β+ =
∂σn /∂[w]+
– the normal relative displacement at critical state, [u]crit
– the dilatancy, ω+ , corresponding to the post-peak phase: ω+ = ∂[u]/∂[w]+
To begin with the formulation of an incremental model, the monotonic CNL and CNS
interface shear tests are considered. Figure 3.6 typically shows the values of ηpeak , [w]η peak ,
σn peak and [u]crit derived from monotonic CNL and CNS tests on dense sand. These key
parameters on loose sand are evaluated in the same way.
On dense sand, ηpeak obviously decreases as a function of σn and k. The effect of k seems to
have more influence on ηpeak with low initial stress than high initial stress. σn peak increases
as a function of σn and k. [w]η peak also increases as a function of σn , whereas the effect
of k seems to have no influence on [w]η peak . An increase of k and σn totally or partially
prevents the change in volume, and consequently leads to the reduction rate of [u]crit ).
0.85 800
ID0 | 90 % ; rough plate
0.83 600
Kpeak = Wpeak/Vn (-)
k = 0 kPa/mm
k = 1000 kPa/mm
Vpeak (kPa)
0.81 k = 5000 kPa/mm

400
0.79
200 k = 0 kPa/mm
0.77 k = 1000 kPa/mm

k = 5000 kPa/mm
0.75 0
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
0.8 0.20
0.6 0.15
[w]K peak (mm)
[u]crit (mm)
0.4 0.10
k = 0 kPa/mm
k = 1000 kPa/mm
k = 5000 kPa/mm
0.2 0.05 k = 0 kPa/mm
k = 1000 kPa/mm
k = 5000 kPa/mm
0 0.00
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
Figure 3.6: Typical peak and critical values of key parameters on dense sand (ID0 ≈ 90%),
rough surface with different values of k
164
8
D0 = cK/c[w]0 (-) 6
4
k = 0 kPa/mm
k = 1000 kPa/mm
k = 5000 kPa/mm
2
0
0 100 200 300 400
Vn (kPa)
0
E0 = cVn/c[w]0 (-)
-100
-200
-300
k = 0 kPa/mm
k = 1000 kPa/mm
k = 5000 kPa/mm
-400
0 100 200 300 400
Vn (kPa)
-10
E+ = cVn/c[w]+ (-)
-20
-30
-40
k = 0 kPa/mm
-50 k = 1000 kPa/mm
k = 5000 kPa/mm
-60
0 100 200 300 400
Vn (kPa)
Figure 3.7: Typical moduli of key parameters on dense sand (ID0 ≈ 90%), rough surface
with different values of k
165
The other figures which deserve to mention at this point, for describing some of the major
parameters which are to properly be interpreted in order to reach the satisfactory model,
are Figure 3.7 and 3.8. These two figures also show the initial slope (the initial moduli and
dilatancy e.g. ∂η/∂[w]0 , ∂σn /∂[w]0 and ∂[u]/∂[w]0 ) as well as the slope corresponding to
the post-peak phase (∂σn /∂[w]+ and ∂[u]/∂[w]+ ).
Considering before-peak phase, ∂η/∂[w]0 decreases as function of σn and k. As can be
observed in CNS tests, σn decreases at the beginning of the test to overcome the contractive
behaviour and the rate of ∂σn /∂[w]0 increases as a function of σn and k. Unlike ∂σn /∂[w]0 ,
∂[u]/∂[w]0 has a different tendency in which an increase of k totally or partially prevents
the volume variation.
After reaching the peak value, the stress ratio decreases (softening behaviour) and subse-
quently evolves to the critical state. As can be observed in the previous chapter, the interface
shear tests can be concluded to evolve the same critical stress ratio (ηcrit = τcrit /σn ≈ 0.566
on rough plate) on both densities. The imposed normal stiffness has no influence on the
critical stress ratio. The rate of ∂σn /∂[w]+ still increases as a function of σn and k. The
trend of dilatancy rate at post peak phase is similar to that at before peak phase.
These basic key parameters are then formulated as a simple function of k and σn for each
density. Table 3.1 shows the summary of peak and critical values of the key parameters on
both densities. The moduli and the dilatancy of the interface (before and post-peak phases)
are then presented on Table 3.2.
0 0.000
-0.04 -0.004
Z+ = c[u]/c[w]+ (-)
Z0 = c[u]/c[w]0 (-)
-0.08 -0.008
-0.12 -0.012
k = 0 kPa/mm k = 0 kPa/mm
k = 1000 kPa/mm k = 1000 kPa/mm
-0.16 -0.016
k = 5000 kPa/mm k = 5000 kPa/mm
-0.2 -0.020
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 3.8: Key parameters for the description of dilatancy on dense sand
166
Table 3.1: Summary of the peak and critical values of the key parameters on rough plate
Key parametrs Function
ID0 ≈ 30 % ID0 ≈ 90 %
ηpeak = a· exp(b.k)
a -8E-05σn + 0.6477 -9E-05σn - 0.8127
b 6E-08σn + 3E-07 7E-09σn - 4E-06
[w]η peak = a· exp(b.k)
a 0.0014σn + 0.5836 6E-04σn + 0.3796
b -9E-08σn - 4E-05 1E-08σn - 5E-06
σn peak = a·k 2 + b·k + c
a 1E-08σn - 2E-07 2E-08σn - 1.3E-05
b -1E-04σn - 0.0036 -1E-04σn + 0.134
c σn σn
[w]σn peak = a· exp(b.k)
a 0.0027σn + 0.5105 0.0011σn + 1.225
b 2E-08σn + 3E-06 -9E-09σn + 9E-06
[u]crit = a·k 2 + b·k + c
a -7E-12σn - 3E-09 -4E-12σn + 4E-09
b 7E-08σn - 2E-05 4E-08σn - 4E-05
c -1E-04σn - 0.0654 -2E-04σn + 0.1753
• Analytical formulation of the base paths

In this study, the analytical formulations for the base paths are then presented. These
formulations were first described by Boutrif [1993] and Garnica-Anguas [1993]. The inter-
polations of the base paths were made from the initial state and the current state of the
interface.
Each of these base paths can analytically be defined from the initial state of the interface
which is characterized by an initial density (ID0 ), initial tangential relative movement ([w]
= 0), an initial normal stress (σn ), and from the current state due to a certain number of
parameters which have been defined in several ways.
For the interface direct shear tests whether in CNL or CNS conditions, the analytical
formulation of the base paths are presented in the space of the incremental response at
different stages of shear loading. These paths can be used to assess the behaviour of the
law in all cases of possible loading. In each phase, analytical formulations of the paths are
written in the function of [w] (the intergration step). Several interpolation functions (e.g.
cubic polynomial, exponential, etc.) are then described. In τ /σn - [w], σn - [w] and [u] − [w]
diagrams, the interpolation functions are given by the following:
167
Table 3.2: Summary of the moduli of the key parameters on rough plate
Key parametrs Function
ID0 ≈ 30 % ID0 ≈ 90 %
α0 (∂η/∂[w]0 ) = a· exp(b.k)
a -0.0027σn + 2.6132 -0.0076σn + 6.0382
b 5E-08σn + 5E-06 4E-08σn - 4E-05
β0 (∂σn /∂[w]0 ) = a·k 2 + b·k
a 4E-08σn + 7E-06 7E-09σn + 7E-06
b -3E-04σn - 0.0816 -2E-04σn - 0.056
β+ (∂σn /∂[w]+ ) = a·k 2 + b·k
a 3E-09σn + 2E-07 4E-09σn + 6E-07
b -2E-05σn - 0.004 -3E-05σn - 0.00875
ω0 (∂[u]/∂[w]0 ) = a·k 2 + b·k + c
a -2E-12σn - 3E-10 -4E-12σn - 7E-10
b 5E-08σn + 9E-06 4E-08σn + 7E-06
c -2E-04σn - 0.129 -3E-04σn - 0.0647
ω+ (∂[u]/∂[w]+ ) = b·k + c
b 5E-09σn + 8E-07 5E-09σn + 6E-07
c -2E-05σn - 0.00612 -2E-05σn - 0.005
• Before-peak phase ; [w] ≤ [w]peak

In this phase, with the peak values and initial moduli, inluding the initial dilatancy, all of
the interface direct shear paths can be written in the same formula as;
!
α0 ηpeak
τ /σn = 2 −2 3 · [w]3
[w]η peak [w]η peak
! (3.2)
α0 ηpeak
+ −2 +3 2 · [w]2 + α0 · [w]
[w]η peak [w]η peak
!
β0 σpeak − σn
σn = −2 · [w]3
[w]2σ peak [w]3σ peak
! (3.3)
β0 σpeak − σn
+ −2 +3 · [w]2 + β0 · [w] + σn
[w]σ peak [w]2η peak
168
!
ω0 [u]crit
[u] = 2 −2 3 · [w]3
[w][u]crit [w][u]crit
! (3.4)
ω0 [u]crit 2
+ −2 +3 2 · [w] + ω0 · [w]
[w][u]crit [w][u]crit
where:
[w]η peak and [w]σ peak are the shear displacement corresponding to ηpeak and σpeak , respec-
tively.
[w][u] crit is the shear displacement corresponding to [u]crit which [w]σ peak = [w][u] crit .
• Post-peak phase ; [w] > [w]peak

The path of τ /σn ([w]) which shows the reduction from peak to critical condition at large
displacement is written as:

τ
τ /σn =
σn crit
" # (3.5)
τ τ ln(0.02) 2
+ − · exp − · ([w] − [w]η peak )
σn peak σn crit (3[w]η peak )2
Because the feature of σn ([w]) is similar to [u]([w]), these two paths are then written in the
same formula:
σn = σpeak + β+ · ([w] − [w]σ peak )

(3.6)

[w]σ peak 2.9957
− β+ · · 1 − exp ([w] − [w]σ peak ) ·
2.9957 [w]σ peak

[u] = [u]crit + ω+ · [w] − [w][u]crit
[w][u]crit (3.7)

2.9957
− ω+ · · 1 − exp ([w] − [w][u]crit ) ·
2.9957 [w][u]crit
The integration of the interface law requires that the base paths, whether at constant
normal stress (CNL) or constant normal stiffness (CNS), have some ”coherences” in their
formulation. The detail of these formulas is provided in Appendix A.
169
Figure 3.9 shows a comparison between the experimental results and the model predictions
of monotonic CNL tests with σn = 120 kPa on both densites. The comparisons of experi-
mental results and model predictions with different values of initial normal stress at CNL
on loose and dense sands are presented in Figure 3.10 and 3.11, respectively.
1 1
ID0 | 30 % ; CNL ; Vn = 120 kPa ID0 | 90 % ; CNL ; Vn = 120 kPa
0.8 0.8
experimental
K WVn
K WVn
model
0.6 0.6
0.4 0.4
experimental
model
0.2 0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
160 160
120 120
experimental experimental
W (kPa)
W (kPa)
model model
80 80
40 40
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
0.05 0.4
0
experimental 0.3
model experimental
-0.05
[u] (mm)
model
[u] (mm)
0.2
-0.1
0.1
-0.15
0
-0.2
-0.25 -0.1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
Figure 3.9: Comparison between experimental data and the model predictions, σn = 120
kPa, CNL tests with rough plate : (left) loose sand (ID0 ≈ 30%) ; (right) dense sand
(ID0 ≈ 90%)
170
In case of CNS test, Figure 3.12 shows the comparison between the experimental tests and
model predictions of monotonic CNS tests with σn0 = 100 kPa, k = 1000, 2000 and 5000
kPa/mm on loose and densed sands, respectively. Results from the model predictions with
various conditions, which cover the main factors influencing the interface behaviour under
CNS condition (e.g. ID0 , σn0 and k), are also shown in Figure 3.13. The model includes
the features of interface model formulated in terms of the incremental stresses and of the
incremental relative displacements. The capability of the proposed model to reproduce the
experimental observations in laboratory is satisfactory. It is also found that the proposed
model can well reproduce the interface behaviour, especially in CNS condition in which
the degradation of normal stress associated with shear stress can well be described at large
tangential displacement. All features of interface model are shown in Appendix B
300
ID0 | 30 % ; CNL
Vn = 60 kPa
ID30_sn60
250 Vn = 120 kPa
ID30_sn120
Vn = 310 kPa
ID30_sn310
200 B
model
W (kPa)
150
100
50
0
0 1 2 3 4 5 6
[w] (mm)
Figure 3.10: Comparison between experimental data and the model predictions, CNL tests
on loose sand (ID0 ≈ 30%) with rough plate
300
ID0 | 90 % ; CNL
Vn = 60 kPa
ID90_sn60
250 Vn = 120 kPa
ID90_sn120
Vn = 310 kPa
ID90_sn310
200 model
m_60
W (kPa)
150
100
50
0
0 1 2 3 4 5 6
[w] (mm)
Figure 3.11: Comparison between experimental data and the model predictions, CNL tests
on dense sand (ID0 ≈ 90%) with rough plate
171
100 400
ID0 | 30 % ; CNS ; ID0 | 90 % ; CNS ;
Vn0 = 100 kPa Vn0 = 100 kPa
k== 1000
1000kPa/mm
kPa/mm k = 1000 kPa/mm
80 k = 2000 kPa/mm
k== 2000
2000kPa/mm
kPa/mm 300
k== 5000
5000kPa/mm
model
1000k
model
1000k
W (kPa)
60
W (kP)
200
40
100
20
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
140 800
ID0 | 30 % ; CNS ; ID0 | 90 % ; CNS ;
120 Vn0 = 100 kPa Vn0 = 100 kPa
k== 1000
1000kPa/mm
kPa/mm kk == 1000
1000 kPa/mm
kPa/mm
k== 2000
2000kPa/mm
kPa/mm 600 kk == 2000
2000 kPa/mm
kPa/mm
100 k== 5000
5000kPa/mm
kPa/mm kk == 5000
5000 kPa/mm
kPa/mm
model
1000k model
1000k
Vn (kPa)
Vn (kPa)
80
400
60
40
200
20
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
120 400
ID0 | 30 % ; CNS ; ID0 | 90 % ; CNS ;
Vn0 = 100 kPa Vn0 = 100 kPa
kk == 1000
1000 kPa/mm
kk == 2000
2000 kPa/mm
kPa/mm 300 k = 2000 kPa/mm
kk == 5000
5000 kPa/mm
kPa/mm kk =
= 5000
5000 kPa/mm
kPa/mm
80 model model
1000k
W (kPa)
1_1000k
W (kPa)
200
40
100
0 0
0 40 80 120 160 0 100 200 300 400 500
Vn (kPa) Vn (kPa)
Figure 3.12: Comparison between experimental data and the model predictions, σn = 100
kPa, k = 1000, 2000 and 5000 kPa/mm with rough plate : (left) loose sand (ID0 ≈ 30%) ;
(right) dense sand (ID0 ≈ 90%)
172
250 400
ID0 | 30 % ; CNS ; ID0 | 90 % ; CNS ;
k = 5000 kPa/mm k = 1000 kPa/mm
Vn0==6060kPa
sn kPa Vn0==60
60kPa
kPa
200 sn
Vn0==100
sn 100kPa
kPa Vn0==100
100kPa
kPa
300 sn
V = 310 kPa
snn0= 300 kPa Vn0==300
sn 310kPa
kPa
B
model B
model
W (kPa)
W (kPa)
150
200
100
100
50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
400 700
ID0 | 30 % ; CNS ; ID0 | 90 % ; CNS ;
k = 5000 kPa/mm k = 1000 kPa/mm
600 Vn0
V
snn0==60
60kPa
kPa sn = 60=kPa
60 kPa
V
snn0==100
100kPa
kPa Vn0
sn = kPa
= 100 100 kPa
300
V
snn0==300
310kPa
kPa 500 Vn0 = 310 kPa
sn = 300 kPa

model

model
Vn (kPa)
Vn (kPa)
400
200
300
200
100
100
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
300 400
ID0 | 30 % ; CNS ;
ID0 | 90 % ; CNS ;
k = 5000 kPa/mm
k = 1000 kPa/mm
V
snn0==60
60kPa
kPa
Vn0==60
sn 60kPa
kPa
Vn0==100
100kPa
kPa
sn 300 Vn0==100
sn 100kPa
kPa
Vn0==300
310kPa
kPa
sn Vn0==300
sn 310kPa
kPa
200 B
model
B
model
W (kPa)
W (kPa)
200
100
100
0 0
0 100 200 300 400 0 100 200 300 400 500
Vn (kPa) Vn (kPa)
Figure 3.13: Comparison between experimental data and the model predictions, σn0 = 60,
100 and 310 kPa with rough plate : (left) k = 5000 kPa/mm on loose sand (ID0 ≈ 30%) ;
(right) k = 1000 kPa/mm on dense sand (ID0 ≈ 90%)
173
3.3 Cyclic tests
A constitutive model for the granular material subjected to a large number of cycles pro-
posed by Wichtmann [2005], Wichtmann et al. [2005] and Niemunis et al. [2005] is devoted
to the explicit method in which the number of cycles is considered instead of time. In the
domain of soil-structure interface, whether under constant normal stress (CNL) or constant
normal stiffness (CNS), a general formulation is expressed as:
d [u] = d [u]e + d [u]cm (3.8)
where d [u]e describes the elastic part of which the increment of elastic displacements is
related to that of stresses according to
d [u]e = De−1 dσ (3.9)
where De is the elasticity matrix expressed as:

k 0
De = t (3.10)
0 kn
While the viscoplastic part a widely-used viscoplastic formulation, so-called Perzyna model
Perzyna [1966], is expressed as
∂G
d [u]cm = Γ hΦ(F )i dN (3.11)
∂σ
in which a fluidity parameter (Γ) and the function of hΦ(F )i describe the intensity of d [u]cm .
A plastic potential G = G(σ) of which ∂G∂σ describes the direction of d [u]cm .
During cyclic tests the variables measured in each test are, in addition to the components
of the stress vector (σcm , τcm ):
• The mean cyclic relative displacement vector, normal [u]cm and tangential [w]cm com-
ponents.
• The initial normal and tangential stiffnesses of the soil-structure interface, kni and kti .
• The strength of sand-interface structure due to one great cycle of shear stress to failure
after cyclic loading phase.
To clarify the interface behaviour under cyclic loading, a representation of stress plane (σn ,
τ ) is useful, with the mean cyclic stress vector (σcm , τcm ), and the mobilisation of mean cyclic
stress ratio (ηcm = στcmcm
), cyclic amplitude (∆η = σ∆τcm
). The remarkable statements involving
the characteristic (δcar ), critical (δcrit ) and peak (δpeak ) angles, or their respective tangents
(ηcar , ηcrit and ηpeak ), depending on the initial density, and their opposite counterparts are
also described. These parameters are derived from monotonic direct shear tests.
174
Figure 3.14 shows the behaviour of the soil-structure interface under cyclic loading which
is characterized into three main poles (in terms of the stress state). In this figure, the pole
here means that the dilating / contracting character is maximum on the level of this pole
(identified in term of ηcm ), and this character decreases in the vicinity of the pole, when
the stress state goes away from this one. It can be distinguished as:
τ ηpeak
ηcrit
ηcar
1/2ηcar
σn
-1/2ηcar
contraction
-ηcar
very contraction
-ηcrit
dilation or contraction
-ηpeak
Figure 3.14: Behaviour of sand-structure interface under cyclic loading, poles in the stress
plan
• 2 poles showing slowly (in term of number of cycles) very contractive domain, regard-
less of the initial density of the sand, and the amplitude of the cycles, for ηcm close to
± 12 ηcar .
• 1 pole showing slowly (in term of number of cycles) less contraction, regardless of the
initial density of the sand, and the amplitude of the cycles, for ηcm close to 0.
• 2 poles driving rapidly (in term of number of cycles) to critical state, regardless
of the initial density of the sand, and the amplitude of the cycles, for ηcm close to
±ηpeak . This behaviour is always dilative in the case of the dense sand, it is dilative or
contractive in the case of the loose sand, according to the initial density, in comparison
with the critical density.
Besides, an increasing amplitude of the cycles amplifies the observed phenomenon.
In shear stress-controlled cyclic tests, the application of the mean cyclic stresses provides
an accumulation of displacements. At first, the model is proposed to properly and simply
describe the mean cyclic displacements ([u]cm and [w]cm ), from the initial state in density
(ID0 ≈ 30 or 90 %) and for the maximum number of cycles (N = 10000) carried out in
laboratory, possibly by extrapolating in a higher number of cycles (N > 20000). However,
175
it will be necessary to question the parameters of cyclic interface behaviour. The current
density is the first one of these parameters, but it seems that the initial density before
cycles is also relevant, as far as the grains have the difference in freedom from a low density
or from a high density, except passing through the critical state (mobilizing in the same
critical-density).
There are three main steps in this approach, making extensive use of successive and pro-
gressive interpolations, which consist of:
• modeling [u]cm path at constant normal stress (CNL)
• modeling [w]cm path at constant normal stress (CNL)
• processing the imposed constant normal stiffness (CNS) paths, still contracting, and
leading to the critical state after a certain number of cycles (as ηcm increases from
ηcm0 , as a correction with respect to the paths at constant normal stress)
It can be noted that the first two steps are in the identification phase whereas the last one
is in the validation phase.
3.3.1 Modeling [u]cm path at constant normal stress (CNL)
A logical first step of modeling [u]cm path is to reach the density (specific weight, γd ) of
the soil-structure interface to connect to the normal relative displacement ([u]cm ) observed.
Thereby, this approach allows the difference between the initial density and the final denstiy
during cyclic loading by considering the normal relative displacement to be interpreted
according to:
!
1
γd = γd0 σn [u]cm
(3.12)
1+ t
where
γd is the current specific weight of interface.
γd0 σn is the initial specific weight of interface under an applied normal stress.
[u]cm is the mean cyclic normal displacement.
t is the interface thickness depending on the initial density (t = 12D50 for ID0 ≈ 30%, and
t = 10D50 for ID0 ≈ 90%).
To validate the model, the global effect ηcm subjected to the attractions of various poles,
on the specific weight (γd ) to 10,000 cycles, is first treated. Then the influence of cyclic
amplitude (∆η = ∆τ /σn ) on the specific weight and finally the influence of the number of
cycles on the evolution of the specific weight at 10,000 cycles are described.
• The global effect of ηcm
176
Let us examine at first the position of the cyclic loading for each reference of normal stress
(σn = 60, 120 and 310 kPa). Figure 3.15 shows the overview of cyclic CNL tests performed
in this study. The tests with high values of ηcm (i.e., ”LH” or ”DH”) are partly positioned
below and partly beyond the characteristic state (contracting and dilating zone under cycles,
respectively). As such, they lead towards the critical state.
The global effect of the cyclic loading depends mainly on the trends identified in five poles
(5 nodes in terms of interpolation) mentioned above:
ηcm = (−ηpeak , −η∗ , 0, η∗ , ηpeak ) (3.13)
ηcar
where η∗ = 2
From these discrete data, the behaviour of soil-structure interface will be generated. Ini-
tially, [u]cm as a function of the number and characteristics of cycles, for any value of ηcm ,
by the interpolation of the behaviour on these discreet data will be described.
• Constitutive interpolations (weight functions)
Two interpolation functions for describing the principal behaviour of interface are
chosen, with regard to ηcm :
1. Term of Wc , ”contraction” : there are five important conditions of the contrac-
tion:
– For ηcm = 0 → Wc = ā with 0 < ā < 1 (contacting for ηcm = 0)
– For |ηcm | = η∗ → Wc = 1
– For |ηcm | = ηpeak → Wc = 0
– For |ηcm | = η∗ → Wc′ = 0
– For |ηcm | = ηpeak → Wc′ = 0
177
1.2 1_10LL60
2_10DL60
3_20LL60
4_20DL60
Kmax , Kmin (-) 0.8 5_10LM60
6_10DM60
7_20LM60
8_20DM60
0.4 9_10LH60
10_10DH60
11_20LH60
12_20DH60
0.0 peak_dense
peak_loose
critical state
characteristic state
-0.4
0 5 10 15 20
N° test
1.2
1_10LL120
2_10DL120
3_10LM120
0.8
4_10DM120
Kmax , Kmin (-)
5_20LM120
6_20DM120
0.4
7_10LH120
8_10DH120
peak_dense
0.0 peak_loose
critical state
-0.4
0 5 10 15 20
N° test
1_10LL310
1.2
2_10DL310
3_20LL310
4_20DL310
5_10LM310
0.8
6_10DM310
Kmax , Kmin (-)
7_20LM310
8_20DM310
9_40LM310
0.4 10_40DM310
11_10LH310
12_10DH310
13_20LH310
0.0 14_20DH310
15_40LH310
16_40DH310
peak_dense
-0.4 peak_loose
0 5 10 15 20 critical state
N° test
Figure 3.15: Position of the cycles with regard to the remarkable parameters on cyclic
interface CNL direct shear tests
178
where Wc (see Appendix C.1 for details) is given as:
3
e 4 η
Wc = ηcm + cm (f + e (ηpeak + η∗ ))
4 3 (3.14)
2
ηcm
+ (eηpeak η∗ − f (ηpeak + η∗ )) + ηcm f ηpeak η∗
2
the parameters e and f are expressed as:

" #
4 (1 − ā)(3η∗ − ηpeak ) ā(3ηpeak − η∗ )
e= + 2 (3.15)
(ηpeak − η∗ )3 η∗2 ηpeak
" #
2 (1 − ā)ηpeak (2η∗ − ηpeak ) āη∗ (2η∗ − ηpeak )
f =− + 2 (3.16)
2. Term of Wd , ”dilation” : similarly, 5 conditions imply the following:

– For ηcm = 0 → Wd = 0
– For |ηcm | = η∗ → Wd = 0
– For |ηcm | = ηpeak → Wd = 1
– For ηcm = 0 → Wd′ = 0
– For |ηcm | = ηpeak → Wd′ = 0
The expression of Wd is:
2
Wd = eηcm (|ηcm − η∗ |) (|ηcm | − f ) (3.17)
where
4ηpeak − 3η∗
f = ηpeak (3.18)
3ηpeak − 2η∗
3ηpeak − 2η∗
e=− 3 (3.19)
ηpeak (ηpeak − η∗ )2
This interpolation gives, as examples, the following results (Figure 3.16 and 3.17 for ID0 ≈
30%, σn = 120 kPa and ID0 ≈ 90%, σn = 310 kPa; and ā = 0.30 in both cases, respectively).
179
1.2
-Kpeak -Kcar +Kcar +Kpeak
weight function, W (-)
0.8
Wd
0.4 Wc
Wd + Wc
dilation contraction contraction dilation

-0.4
-0.8 -0.4 0 0.4 0.8
Kcm
Figure 3.16: Interpolation functions on ηcm , ID0 ≈ 30%, σn = 120 kPa, the summation of
Wc + Wd = 1.119 and the minimum of Wd = -0.0904
1.2
0.8
Wd
0.4 Wc
Wc + Wd
-0.4
-1 -0.5 0 0.5 1
Kcm
Figure 3.17: Interpolation functions on ηcm , ID0 ≈ 90%, σn = 310 kPa, the summation of
Wc + Wd = 1.048 and the minimum of Wd = -0.0347
• Interpolation corrections
Unfortunately, it is found that these interpolations still present a defect, because their
summation of Wc and Wd is not equal to unity ∀ηcm . Therefore, some corrections are
necessary to be treated. However, the range of −η∗ < ηcm < η∗ and the range of η∗ < |ηcm |
180
< ηpeak are considered differently.
In the range of −η∗ < ηcm < η∗ , it is necessary that Wd is zero, because ηcm is in the
contraction area, regulated by Wc and ā only.
While in the range of η∗ < |ηcm | < ηpeak , depending on Wc and Wd , it would be satisfactory
if the summation is equal to unity. The error with respect to these requirements still seems
to be quite low, but this should still make a correction to Wc and Wd . Therefore, the so-
called Wc cor and Wd cor will be constructed for the correction of Wc and Wd , respectively.
Two corrective coefficients (”cfc” and ”cfd”) are then described as:
cfd = −0.5 + sgn (|ηcm | − η∗ ) exp(− ||ηcm | − η∗ |)0.2 (3.20)
with
cfc = cfd + 1 (3.21)
where ”cfc” and ”cfd” are the corrective coefficients of contractive and dilative zones, re-
spectively. Figure 3.18 shows these two corrective coefficients as a funcion of ηcm on both
densities.
1.5 1.5
coefficient of weight function, W (-)
coefficient of weight function, W (-)
1 1
0.5 0.5
0 0
-0.5 -0.5
coefd dilation
coefc contraction
-1 -1
-1.5 -1.5
-0.8 -0.4 0 0.4 0.8 -1 -0.5 0 0.5 1
Kcm Kcm
(a) (b)
Figure 3.18: Corrective coefficients of Wc and Wd : (a) in case of loose sand ; (b) in case of
dense sand
Then the correction of Wc cor and Wd cor can be written as:
Wd (1 + cfd)
Wd cor = (3.22)
cfx
Wc cor = Wc − cfc(Wc + Wd cor − 1) (3.23)
181
where ”cfx” is the maximum value of Wd (1 + cfd). The value of ”cfx” can be given in the
following:
0.9089(σn )0.0016

for dense sand
cfx = (3.24)
0.9424(σn )0.0014 for loose sand
As a result, Figure 3.19 and 3.20 show again the interpolation functions for ID0 ≈ 30%, σn
= 120 kPa and ID0 ≈ 90%, σn = 310 kPa in which the summation of Wc cor and Wd cor is
unity.
1.2
0.8
Wd
Wc
Wd + Wc
0.4
Wd cor
Wc cor
Wc+Wd cor
0

-0.4
-0.8 -0.4 0 0.4 0.8
Kcm
Figure 3.19: Interpolation functions on ηcm , ID0 ≈ 30%, σn = 120 kPa
In view of the simulations, these interpolations are effective for ηcm taken away from ηpeak .
On the other hand, ηcm can never exactly approach ηpeak unless ∆η ≈ 0. It is therefore
necessary to consider the position of ηmax in comparison with ηpeak , because, unlike ηcm ,
ηmax can exactly approach ηpeak . The case of ηcm > 0 will be first considered, it will be
necessary to revise ηmax and ηmin later so as to incorporate the case of ηcm < 0 because
ηmin can approach −ηpeak . The correction is obtained by ”amplifying” ηcm in a controlled
manner, becoming ηcm2 in the direction of ηpeak , according to the following strategy;
ηmax
• if ηpeak is very small → ηcm2 ≈ ηcm
ηmax
• if ηpeak → 1 then ηcm2 → ηpeak
Finally, the expression of ηcm2 is given (see Appendix C.2 for details) as:
" 4 3 2 #
33 ηmax 29 ηmax 21 ηmax
ηcm2 = ηcm + (ηpeak − ηcm ) − + − (3.25)
4 ηpeak 2 ηpeak 4 ηpeak
182
This correction is then applied by using ηcm2 instead of ηcm .
1.2
0.8
Wd
Wc
Wc + Wd
0.4
Wd cor
Wc cor
Wc + Wd cor
0
-0.4
-1 -0.5 0 0.5 1
Kcm
Figure 3.20: Interpolation functions on ηcm , ID0 ≈ 90%, σn = 310 kPa
During a cyclic loading, [u]cm results from variations of the specific weight of interface. This
specific weight of interface evolves between its initial value (expressed under the application
of stress) γd0 σn and its value at the end of cyclic loading (Nf ), i.e. Nf = 10000 in the
general case, and Nf < 10000 in the case of a cyclic loading leading to the critical state.
From experimental point of view, the evolution of specific weight ends up with Nf ≤ 10000
(γd σn Nf ) in the general case.
γd σn Nf is then calculated as the weighted value between:
• a maximum value (γdmax σn Nf ) observed in the neighborhood of the very contracting
poles ±η∗ ,
• and a critical value (γdcrit σn ) mostly observed in the neighborhood of the dilating
poles ±ηpeak ,
The weighting coefficients are respectively the functions Wc cor and Wd cor previously de-
scribed as a function of ηcm . And then the specific weight for a given N (γd N ) can be
expressed as:
γd N = γd0 σn + Wc cor (γdmax σn Nf − γd0 σn ) + Wd cor (γdcrit σn − γd0 σn ) (3.26)

• Influence of cyclic amplitude ∆η
Let us consider the influence of ∆η on γdmax σn Nf in the neighborhood of the very contracting
pole (ηcm ≈ ηcar
2 ). A power formulation of γdmax σn Nf is given as:
183
γdmax σn Nf = γd∗ (∆η)n (3.27)
The performance of this formulation for three normal stresses has served as a reference. The
parameters γd∗ and n relative in ηcm ≈ ηcar
2 are represented in Figure 3.21. The analytical
formulations for each density are given by
• for loose sand
γd∗ = 24.4exp(−0.0009σn ) (3.28a)

n = 0.1567exp(−0.0043σn ) (3.28b)
• for dense sand
γd∗ = 19.37exp(−0.00023σn ) (3.29a)

n = 0.0311exp(−0.003σn ) (3.29b)
24 0.16
loose loose
dense dense
22 0.12
Jd max (kN/m3)
exponent, n (-)
y = 24.4e-0.0009x
R2 = 0,9773 y = 0.1567e-0.0043x
R² = 0.9457
20 0.08
18 0.04 y = 0.0311e-0.003x
y = 19.37e-0.00023x R² = 0.7017
R2 = 0,8127
16 0
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
ηcar
Figure 3.21: Parameters γd∗ and n, representation of γdmax σn Nf relative in ηcm ≈ 2
Let us examine now the influence of ∆η on γd σn Nf for the high values of ηcm which does
not allow further testing until N = 10000, because the critical state is reached for a small
number of cycles, with large tangential relative displacement. The specific weight at this
state is then taken into account as γdcrit σn . The evolution between initial specific weight
(after the application of normal stress) and final specific weight in case of high values of
ηcm is represented in Figure 3.22. It is found that the specific weight at critical state
from monotonic tests is slightly lower than that from cyclic tests. As mentioned in the
previous chapter, the thickness of interface deduced from monotonic tests (with a very
good coefficient of correlation) is lower in high initial density (t = 10D50 ) than in low initial
density (t = 12D50 ). The influence of ∆η on γdcrit σn is generally low, but requires further
analysis.
184
17 16.6
Jd Vn (ID0 = 90%)
monotonic CNL
16.5 cyclic CNL
16.2
y = 0.0018x + 15.477
Jd (kN/m3)
16
Jd crit (kN/m3)
R2 = 0,9741
15.5 y = 0,0018x + 15,477 15.8
y = 0.0023x + 15.285
15 R2 = 0,9955
Jd Vn (ID0 = 30%) 15.4
14.5
14 15
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 3.22: (a) Cyclic tests with high values of ηcm , evolution between initial and final
specific weights; (b) Critical specific weights from monotonic and cyclic tests
• Evaluation of the parameter ā characterizing the moderately contractive

pole (ηcm = 0)
The formulation of ā relates γd σn Nf in the vicinity of ηcm = 0 and ηcm ≈ 21 ηpeak . This
parameter mainly influences on the low value of ηcm . The ratio of ā is derived from the
difference of γd σn Nf between ηcm = 0 and ηcm ≈ 12 ηpeak . It is worth mentioning that since
ā is the interpolation coefficient of Wc , the other one (Wd ) is then zero. The variations of
ā as a function of σn can be given as:

0.8329(σn )−0.0423 for loose sand
ā = (3.30)
0.844(σn )−0.0509 for dense sand
• Rate of cyclic pseudo-creep evolution (constant N50% )
The rate of cyclic pseudo-creep evolution is set by the function of the number of cycles
f (N ). Initially, two functions are used to describe this rate (cyclic pseudo-creep) in each of
the poles (contraction and dilation):
N
fc (N ) = (3.31a)
Cte1 + N
fd (N ) = 1 − exp(−Cte2 · N ) (3.31b)
where fc (N ) and fd (N ) are respectively the functions applying to the contractive and
dilative poles. These functions are represented and parametrized in Figure 3.23. As can
be seen, the hyperbolic function (fc ) practically reaches the unity for approximately 5,000
cycles with Cte1 = 670, whereas the exponential function (fd ) practically reaches the unity
for approximately 1000 cycles with Cte2 = 0,003912.
185
1.2 1.2
1.0 1.0
f(N) = 1 - H[S&WHǜ1

f(N) = N/(Cte + N)
0.8 0.8
0.6 0.6
cte = 0,03912
0.4 0.4 cte = 0,003912
cte = 670 cte = 0,0003912
0.2 cte = 6700 0.2
cte = 67000
0.0 0.0
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
N (-) N (-)
(a) (b)
Figure 3.23: (a) Hyperbolic function of the number of cycles f (N ); (b) Exponential function
of the number of cycles f (N )
Regarding the results of cyclic tests, the evolutions of contraction are relatively slow,
whereas the evolutions of dilation (or towards the critical state) are rather fast. This
remark allows the choice of the functions and the constants, according to the values of σn ,
ηcm , etc.
N
The hyperbolic function fc (N ) = Cte1+N seems more properly applicable, because the
direction of the constant is easy to identify. Figure 3.24 shows the direction of Cte from
hyperbolic function. It is clear that the Cte = N50% is appropriate to use.
10000
8000
Nfunction (-)
6000
4000
f(N) = 40%
2000 f(N) = 50%
f(N) = 75%
f(N) = 95%
0
0 2000 4000 6000 8000 10000
N (-)
Figure 3.24: Direction of constant hyperbolic function
186
1600
10LL60
N50% (ID0=90%) = -1.8114Vn + 1118
20LL60
1200
10LL120
N50% (Kcm = 0)
10LL310
20LL310
800 10DL60
20DL60
10DL120
400 10DL310
N50% (ID0=30%) = -1.8114Vn + 1118
20DL310
0
0 100 200 300 400
Vn (kPa)
1600
10LM60
20LM60
10LM120
1200 20LM120
N50% (Kcm ~ 1/2Kpeak)
10LM310
20LM310
N50% (ID0=30%) = -0.5829Vn + 629
40LM310
800
10DM60
20DM60
10DM120
400 20DM120
10DM310
20DM310
N50% (ID0=90%) = -0.6086Vn + 524
40DM310
0
0 100 200 300 400
Vn (kPa)
1000
10LH60
20LH60
800
N50% (Kcm ~ 9/10Kpeak)
10LH120
10LH310
600 40LH310
20LH310
N50% (ID0=90%) = -0.2029Vn + 243
10DH60
400 20DH60
10DH120
10DH310
200
N50% (ID0=30%) = 0.0586Vn + 44 20DH310
40DH310
0
0 100 200 300 400
Vn (kPa)
Figure 3.25: Evaluation of the number of cycles N50% to achieve 50% of [u]cm reached in
Nf (Nf = 10000 for low and medium values of ηcm , less for high value of ηcm )
187
Therefore, it is a question of allocating a constant to each test by considering the rate of
evolution of the test. Figure 3.25 gives an evaluation of the degree of evolution of [u]cm
with N (the number of cycles, N50% , allowing to reach 50% of [u]cm reached in Nf ). From
this degree of evolution, the hyperbolic constant N50% is then be held. The average lines
for each of 3 typical values of ηmax /ηpeak as a functions of σn are then taken into account as
shown in Figure 3.26. And hence a trend of the intercept and slope of these average lines
as a function of ηmax /ηpeak is obtained.
0.5 1200
loose sand
intercept of N50% linear/Vn cm

1000
slope of N50% linear/ Vn cm
loose dense sand

0
dense y = 1.87x ± 1.7136
R2 = 0,9682
800
-0.5
600
y = -875x + 1065.8
-1 R2 = 0,9591
400
y = 1.6085x ± 1.6786
R2 = 0,9243
-1.5
200
y = -1074x + 1134
R2 = 0,9973
-2 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
Kmax/Kpeak Kmax/Kpeak
(a) (b)
Figure 3.26: Slope and intercept of linear approximations of N50% (σn ) as a function of
ηmax /ηpeak
The formulation of the constant N50% is then expressed as:
ηmax ηmax
for loose sand f (N50% ) = (−1074 + 1134) + (1.87 − 1.7136)σn (3.32a)
ηpeak ηpeak
ηmax ηmax
for dense sand f (N50% ) = (−875 + 1065.8) + (1.6085 − 1.6786)σn (3.32b)
ηpeak ηpeak
Therefore, from (3.26) the evolution of specific weight during cyclic loading is given:
γd N = γd σn + f (N50% )Wc cor (γdmax σn Nf − γd0 σn )

(3.33)
+ f (N50% )Wd cor (γdcrit σn − γd0 σn )
Figure 3.27 shows typical prediction of the evolution of specific weight during cyclic loading
for σn = 120 kPa, ID0 ≈ 30% and σn = 310 kPa, ID0 ≈ 90% with ∆τ = 10 kPa. Therefore,
the evolution of the normal displacement [u]cm as a function of N is achieved.
188
18 18
17 17
Jd (kN/m3)
Jd (kN/m3)
16 16
10DL310
10LL120 10DM310
15 10LM120
15
10DH310
10LH120
14 14
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N (-) N (-)
(a) (b)
Figure 3.27: Evolution of specific weight during cyclic loading: (a) σn = 120 kPa, ID0 ≈ 30%
; (b) σn = 310 kPa, ID0 ≈ 90%
3.3.2 Modeling [w]cm path at constant normal stress (CNL)
The evolution of [w]cm depending on the number of cycles is similar to that of the creep
(Figure 3.28): it is the cyclic pseudo-creep. In the case of soil-structure interfaces studied,
three main characteristics can be observed:
• no pseudo-cyclic creep for a very low value of ηcm (i.e. ηcm ≈ 0)
• primary and secondary creep for the medium values of ηcm (i.e. ηcm ≈ η∗ )
• primary and then secondary then tertiary creep for the high values of ηcm (i.e. ηcm
close to ηpeak )
primary secondary tertiary

N
x
deformation (H)
x1 X x2
time (t)
Figure 3.28: The classical view of creep
189
The evolution of [w]cm will be treated by a function of the number of cycles (as described
in the part of [u]cm ). First, the expression of the limits, for medium and high values of
ηcm has to be described. The estimation of the oblique asymptotes of [w]cm (N ), and the
rate of evolution towards these asymptotes will then be treated separately. Figure 3.29
then represents these estimated asymptotes, for medium value of ηcm (so-called secondary
creep).
Let us now separate the slope and intercept of the asymptotes.
• Slope of the asymptote of [w]cm (N ) for the medium values of ηcm
It is here about secondary pseudo-creep. Figure 3.30 represents the slope m of the estimated
asymptotes of [w]cm (N ) on loose and dense sands, with a parameter ∆η. A formulation of
this slope which takes into account, at the same time, the level of the normal stress and the
cyclic amplitude is determined. The effect of stress level on the slope of the asymptote, in
case of loose sand, can be expressed in a power law as:
m[w]cm/cycle = 0.0457(σn )−1.691 (mm/cycle) (3.34)
2.5 1.0
asymptote of [w]cm (mm)
asymptote of [w]cm (mm)
2 10LM60
0.8 10DM60
20DM60
20LM60
10DM120
1.5 10LM120 0.6
20LM120 20DM120
10LM310 10DM310
1 20LM310 0.4 20DM310
40LM310 40DM310
0.5 0.2
0 0.0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N N
(a) (b)
Figure 3.29: Estimation of asymptotes in case of medium value of ηcm : (a) loose sand,
ID0 ≈ 30% ; (b) dense sand, ID0 ≈ 90%
In case of dense sand, the slope of the asymptote is still expressed in a power law:
m[w]cm/cycle = 6E − 05(σn )−0.427 (mm/cycle) (3.35)
Taking into account the influence of ∆η on m[w]cm/cycle in power law, a coefficient c describ-
ing the ratio of m for each ∆η has to be presented. Figure 3.31 shows the coefficient (c) as
a function of ∆η for loose and dense sands. This can be written as:
190
6.E-05 1.2E-05
Cyclic CNL tests ; ID0 y30% y90%
Cyclic CNL tests ;I ID0y90%
D0
5.E-05 1.0E-05
y= 6E-05x-0.427
slope, m (mm/cycle)
slope, m (mm/cycle)
R² = 0.9838
4.E-05 8.0E-06
3.E-05 6.0E-06
y = 0.0457x-1.691
R² = 0.9026
2.E-05 4.0E-06
1.E-05 2.0E-06
0.E+00 0.0E+00
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 3.30: Estimated slope of the asymptote [w]cm (N ) in case of medium value of ηcm :
(a) loose sand, ID0 ≈ 30% ; (b) dense sand, ID0 ≈ 90%
for loose sand c = −9.468∆η 2 + 7.3472∆η − 0.0107 (3.36a)

for dense sand c = 1.5405∆η 2 + 5.9432∆η + 0.0005 (3.36b)
Finally, the slope of the asymptote for the medium values of ηcm can be written as:
for loose sand m[w]cm/cycle = 0.0457(σn )−1.691 (−9.468∆η 2 + 7.3472∆η − 0.0107) (3.37a)
for dense sand m[w]cm/cycle = 6E − 05(σn )−0.427 (1.5405∆η 2 + 5.9432∆η + 0.0005) (3.37b)
1.6 2.5
Coefficient of slope ; ID0 y30% Coefficient of slope ; ID0 y90%
1.2 2
coefficient, c (-)
coefficient, c (-)
0.8 1.5
y = -9.468x2 + 7.3472x - 0.0107
R² = 0.951
0.4 1
y = 1.5405x2 + 5.9432x + 0.0005
R² = 0.983
0.0 0.5
-0.4 0
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
'K 'K
(a) (b)
Figure 3.31: Coefficient ”c” of the slope of asymptote in case of medium value of ηcm : (a)
loose sand, ID0 ≈ 30% ; (b) dense sand, ID0 ≈ 90%
191
It is worth noting that m[w]cm/cycle 6= 0 for ∆η = 0, which is not physical, but very low,
however, this results from the approximation mode.
• Intercept of the asymptote of [w]cm (N ) for the medium values of ηcm
Similarly to the slope of asymptote, the estimated intercept of the asymptote (b[w]cm ) as a
function of normal stress level can be expressed in both cases:
for loose sand b[w]cm = 20.019(σn )−0.6281 (3.38a)

for dense sand b[w]cm = 2.5742(σn )−0.3584 (3.38b)
A coefficient c taking into account the influence of ∆η on b[w]cm can be given in power law
as:
for loose sand c = 1.6359∆η 0.2944 (3.39a)

0.287
for dense sand c = 1.6676∆η (3.39b)
And then the intercept of the asymptote for the medium values of ηcm can be written as:
for loose sand b[w]cm = 20.019(σn )−0.6281 (1.6359∆η 0.2944 ) (3.40a)

for dense sand b[w]cm = 2.5742(σn )−0.3584 (1.6676∆η 0.287 ) (3.40b)
• Evolution of [w]cm (N ) with N for the medium values of ηcm
The evolution of [w]cm (N ) is described by using a hyperbolic formulation. The oblique
asymptote is formulated according to the slope and intercept of the estimation.
y1 ax b
b y y1 y 2
0 x
bx
y2
(C x )
b
Figure 3.32: Evolution of [w]cm (N ) with N , linear + hyperbolic functions
192
[w]cm-asymp = m[w]cm/cycle · N + b[w]cm (3.41)
The evolution of [w]cm as a function of N is written as:

N
[w]cm = [w]cm-asymp − b[w]cm · 1 − (3.42)
Cte + N
or it can be rewritten in
N
[w]cm = m[w]cm/cycle · N + b[w]cm · (3.43)
Cte + N
The constant Cte is comparable to N50% as mentioned in the rate of cyclic pseudo-creep
evolution. In fact, the evolution of [w]cm is slower than that of [u]cm . The set of constant
Cte is decided to be:
Cte = 2N50% (3.44)

• Modeling of [w]cm for high value of ηcm
In case of tertiary pseudo-creep (high value of ηcm ), the estimation of slopes and intercepts
is more difficult and subjective than in the case of the secondary pseudo-creep (medium
value of ηcm ). In addition, the evolution of hyperbolic formulation as a function of N is no
longer appropriate because the curve [w]cm (N ) contains an inflexion point (at the beginning
of the so called tertiary creep).
In the hypothesis of high value of ηcm , the model parameters of cyclic CNL tests mainly
depend on σn , ηcm and ∆η.
Let us see how this evolution will be modeled. Two linear formulations (as shown in Figure
3.33) are first chosen as:
y1 = y0′ · x (3.45a)
y2 = a · x + c (3.45b)
These two linear formulations are based on:

• a and b the slope and intercept of the asymptote
• y0′ the slope passing through the origin
• horizontal interpolation between these two lines
193
[ w]cm , y
y1
1 y2
y '0
N
x1 X x2 x
a
b 1
Figure 3.33: Evolution of [w]cm (N ) with N at high value of ηcm , linear function
For a given y , the formulation can commonly be written as:
N = H1 x1 + H2 x2 (3.46)
y1 y2 −b
where x1 = y0′
, x2 = a and 0 ≤ H1 , H2 ≤ 1.
The function H1 and H2 are given in the following:
H1 = exp(−Ay 2 ) (3.47a)
H2 = 1 − exp(−Ay 2 ) (3.47b)
where ”A” is a parameter describing the inflexion point, so-called ”[w]cm inf ”, which can be
expressed as:
1
A= 1/2
(3.48)
[w]cm inf
Substitution of H1 and H2 into 3.46 gives:
y y−b
exp(−Ay 2 ) + 1 − exp(−Ay 2 )

X= ′ (3.49)
y0 a
or it can be rewritten in [w]cm − N form:
[w]cm [w]cm − b
exp(−A[w]2cm ) + 1 − exp(−A[w]2cm )

N= ′ (3.50)
y0 a
194
To be able to use two linear formulations, the determination of a, b and y0′ has to be de-
scribed. Before handling these parameters, the so-called ∆ηmax which describes the feasible
condition that the tests can be carried out is considered. ∆ηmax principally depends on ηcm
which can be expressed as (see Appendix C.3 for details):
ηpeak − |ηcm |
∆ηmax = 2 (3.51)
1 + ηpeak · |ηcm |
From the above expression, it turns out that

0 if |ηcm | → ηpeak
∆ηmax =
2ηpeak if |ηcm | → 0
Then, the current parameter p for describing these tests are proposed.
Let p be:
• the slope and intercept of the asymptote, a and b
• or the slope passing through the origin, y0′
• or the inflexion point, A
p max 'K
Kcm o K peak
Kcm Cte
Kcm o 0
'K
'Kmax 2K peak
Figure 3.34: Limit of pmax (∆η) for a given |ηcm |
The determination of these parameters can be represented in p − ∆η plane for each level
of ηcm as shown in Figure 3.34. It can be noted that the curves p(∆η, |ηcm |) are limited
by the curve pmax (∆η), corresponding to ∆η = ∆ηmax for a given |ηcm |. The expression is
then given as:

∆η ∆η
p = pmax · 2− (3.52)
∆ηmax ∆ηmax
195
And then the maximum feasible conditions of a, b and y0′ (later called amax , bmax and y0′ max ,
respectively), corresponding to ηcm are determined for each level of σn :
b1 c 1
amax = a1 ∆ηmax + b1 − (3.53)
c1 − ∆ηmax
b2 c 2
bmax = a2 ∆ηmax + b2 − (3.54)
c2 − ∆ηmax
b3 c 3
y0′ max = a3 ∆ηmax + b3 − (3.55)
c3 − ∆ηmax
where ai , bi , and ci are the parameters which only depend on σn .

Figure 3.35 shows typical parameters ai , bi , and ci derived from the estimation in case of
σn = 310 kPa on dense sand. Although certain points are beyond the maximum criterion
(e.g. ai ), the evolution of theses parameters coresponding to ηcm is still reliable.
However, to obtain the proper formulation, the elaboration of these parameters, by consid-
ering the size of the cycle in term of the ratio of ∆η/ηmax and ηpeak /∆η, is still required. The
multiplicative factors (Ca max , Cb max and Cy′ 0 max ) are then taken into account, depending
on the initial density, and these factors can be given as the following:
• in case of loose sand

∆η 1 ηpeak
Ca max = 1− 3.8592 (3.56a)
ηmax 100 ∆η

∆η 1 ηpeak
Cb max = 1− −15.096 − 19.245 (3.56b)
ηmax 100 ∆η

∆η 1 ηpeak
Cy′ 0 max = 1− −20 + 1.5 (3.56c)
ηmax 100 ∆η
• in case of dense sand

∆η 1 ηpeak
Ca max = 1− −30 + 4 (3.57a)
ηmax 100 ∆η

∆η 1 ηpeak
Cb max = 1− −53.177 −5 (3.57b)
ηmax 100 ∆η

∆η 1 ηpeak
Cy′ 0 max = 1− 0.886 + 1.931 (3.57c)
ηmax 100 ∆η
196
0.04
0.034( 0.025)
a max 0.01263'K max 0.034
0.025 'K max
0.03
a (mm/cycle)
10DH310
0.02 20DH310
40DH310
Kcm = 0,5341
etacm mod
0.534 (model)
Kcm = 0,5646
etacm 0.565 (model)
mod
0.01 Kcm = 0,6518
etacm 0.652 (model)
mod
Kcm = 0,6828
etacm 0.683 (model)
mod
0.00
0.00 0.20 0.40 0.60 0.80
'K
0.0
10DH310
-1.0 20DH310
40DH310
Kcm = 0,5341
etacm mod
0.534 (model)
-2.0 Kcm = 0,5646
etacm 0.565 (model)
mod
Kcm = 0,6518
0.652 (model)
b (mm)
etacm mod
Kcm = 0,6828
etacm
0.683 (model)
mod
-3.0
-4.0
5.5872( 0.025)
bmax 3.512'K max 5.5872
0.025 'K max
-5.0
0.00 0.20 0.40 0.60 0.80
'K
0.005
0.004
0.0042( 0.025)
y ' 0 max 0.00265'K max 0.0042
y'0 (mm/cycle)
0.025 'K max
0.003
10DH310
0.002 20DH310
40DH310
KK = 0.534 (model)
cm = 0.534
etacm
cm 0,5341(model)
mod
KKcm ==0.565
0.565(model)
0.001 etacm
cm 0,5646(model)
mod
KKcm ==0.652
0.652(model)
etacm
cm 0,6518(model)
mod
KKcm ==0.683 (model)
0.683 (model)
cm
etacm 0,6828 mod
\¶ 0 max
0
0.00 0.20 0.40 0.60 0.80
'K
Figure 3.35: Estimation of ai , bi , and ci for high value of ηcm , σn = 310 kPa on dense sand
197
The inflexion points ([w]cm inf ) which mainly depend on σn and the ratio of η∆η
peak
are deter-
mined from experimental data. Figure 3.36(a) shows the typical derivation of [w]cm inf in
case of loose sand. The influence of cyclic amplitude represented by η∆η
peak
on the inflexion
of [w]cm (N ) is also presented in Figure 3.36(b).
∆η
From taking into account the influence of σn and ηpeak , the expression of inflexion points
([w]cm inf ) on both densities can be given as:

1 ∆η
for loose sand; [w]cm inf = (−0.499lnσn + 3.5571) −0.0364 + 1.031 (3.58a)
0.0536 ηpeak

1 ∆η
for dense sand; [w]cm inf = (−0.731lnσn + 5.624) −0.0892 +1 (3.58b)
0.0389 ηpeak
2.0 2.0
experimental
'KKpeakmod
Dh/hpic 0,0536
0.054 (model)
1.6 1.6 'KKpeakmod
Dh/hpic 0.137 (model)
0,1370
y = -0.499Ln(x) + 3.5571 'KKpeakmod
0,2658
[w]cm inf (mm)
[w]cm inf (mm)
R2 = 0,8239 'KKpeakmod
0,5317
1.2 1.2
0.8 0.8
0.4 0.4
0.0 0.0
0 100 200 300 400 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 3.36: Typical inflexion of [w]cm (N ) for the high values of ηcm on loose sand
Combining (3.43), called [w]cm 1 , and (3.49), called [w]cm 2 , with the constitutive functions
Wc cor and Wd cor , respectively, the formulation of [w]cm for a given ηcm is written as:
[w]cm = [w]cm 1 (Wc cor ) + [w]cm 2 (Wd cor ) (3.59)
From experimental data, [w]cm at ηcm = 0 is not prominent. This also means that Wc cor
and Wd cor for modeling [w]cm should be zero when ηcm is close to zero. Therefore, Wc cor
and Wd cor additionally require multiplicative factors, which can respectively be written as:
" 8 !#
(η cm − η peak )
Wmc cor = Wc cor exp − 2.42 (3.60a)
ηpeak
Wmd cor = Wd cor [1 − exp(−92.1034ηcm )] (3.60b)
198
3.3.3 Checking of the proposed analytical models for identification
To check the validation of the proposed models, Figure 3.37 - 3.40 show typical comparisons
between experimental observations and analytical model predictions of cyclic CNL tests by
considering the overall factors (e.g. ID0 , σn , ηcm and ∆τ ). Generally, the proposed models
are capable of reproducing the interface behaviour under cyclic loading. In the range of 0
< ηcm < ηcar , the capability of the proposed models is satisfactory. However, in case of ηcm
close to ηpeak , it is difficult to properly reproduce the cyclic interface behaviour. The early
termination of these tests provides a deficient description of the rate of cyclic pseudo-creep
evolution. At high level of ηcm close to ηpeak , during cyclic shearing the model cannot
properly reproduce [u]cm especially on dense sand. All features of interface model in CNL
condition are summarized in Appendix D.
In addition, the proposed models are verified by interpolating wheteher on ID0 , ηcm or σn .
Figure 3.41 first shows the model predictions of cyclic CNL tests on ID0 = 60%, σn =
310 kPa. In this case the medium sand shows approximately contractive behaviour at ηcm
close to ηpeak . It is noted that the sign of ηcm (±ηcm ) only indicates the direction of shear
displacement while this has no influence on normal displacement. The interpolation of σn
= 200 kPa also confirms the validity of the proposed models as shown in Figure 3.42.
0.0 0.0 0
10LL120 10LH120
10LM120
model -0.1 -0.04 model
-0.1 model
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.2 -0.08
-0.2
-0.3 -0.12
-0.3
-0.4 -0.16
-0.4 -0.5 -0.2

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-)
N (-)
0.2 1.6 4
10LL120 10LM120
model model
0.1 1.2 3
[w]cm (mm)
10LH120
[w]cm (mm)
[w]cm (mm)
model
0.0 0.8 2
-0.1 0.4 1
-0.2 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-) N (-)
Figure 3.37: Comparison between experimental data and the model prediction, σn = 120
kPa, cyclic CNL tests on loose sand (ID0 ≈ 30%), ∆τ = 10 kPa with rough plate
199
0.00 0.00 0.12
10DL310 10DM310 10DH310
model model model
-0.02 -0.04 0.08
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.04 -0.08 0.04
-0.06 -0.12 0
-0.08 -0.16 -0.04

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-) N (-)
0.2 0.4 3
10DL310 10DM310
10DH310
model model
0.1 0.3 model
2
[w]cm (mm)
[w]cm (mm)
[w]cm (mm)
0.0 0.2
1
-0.1 0.1
-0.2 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-) N (-)
kPa, cyclic CNL tests on dense sand (ID0 ≈ 90%), ∆τ = 10 kPa with rough plate
0.0 0 0
20LM120 40LM310 40LH310
model model -0.05 model
-0.2 -0.1
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.1
-0.4 -0.2
-0.15
-0.6 -0.3
-0.2
-0.8 -0.4 -0.25

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600 2000
N (-) N (-) N (-)
1.6 0.8 4
20LM120 40LM310 40LH310
model model model
1.2 0.6 3
[w]cm (mm)
[w]cm (mm)
[w]cm (mm)
0.8 0.4 2
0.4 0.2 1
0.0 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600 2000
N (-) N (-) N (-)
and 310 kPa, cyclic CNL tests on loose sand (ID0 ≈ 30%), ∆τ = 20 and 40 kPa with rough
plate
200
0.00 0.00 0.12
20DM120 40DM310 40DH310
-0.04 model -0.04 model
model 0.08
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.08 -0.08
0.04
-0.12 -0.12
0.00
-0.16 -0.16
-0.20 -0.20 -0.04

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 200 400 600 800
N (-) N (-) N (-)
0.8 0.6 3
20DM120 40DH310
40DM310
model
0.6 model model
[w]cm (mm)
[w]cm (mm)
0.4 2
[w]cm (mm)
0.4
0.2 1
0.2
0.0 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 200 400 600 800
N (-) N (-) N (-)
and 310 kPa, cyclic CNL tests on dense sand (ID0 ≈ 90%), ∆τ = 20 and 40 kPa with rough
plate
0.00 0 0.2
10LH310
10DH310
-0.04 0.1
ID0 = 60% , Kcm = +/- 0.65
ID=60%
-0.1
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.08 0.0
-0.12 -0.1
-0.2
10LL310 10LM310
-0.16 10DL310 10DM310 -0.2
ID0 = 60%, Kcm = 0
ID=60% ID0 = 60% , Kcm = +/- 0.30
ID=60%
-0.20 -0.3 -0.3
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 1000 2000 3000 4000 5000
N (-) N (-) N (-)
0.2 0.4 4
0.1 0.2 2
[w]cm (mm)
[w]cm (mm)
[w]cm (mm)
10LM310
0.0 0.0 0 10LH310
10DM310 10DH310
ID0 = 60%, Kcm = +/- 0.30
ID=60% ID0 = 60%, Kcm = +/- 0.65
10LL310 ID=60%
-0.1 10DL310 -0.2 -2
ID0 = 60%, Kcm = 0
ID=60%
-0.2 -0.4 -4
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 1000 2000 3000 4000 5000
N (-) N (-) N (-)
Figure 3.41: Interpolation of cyclic CNL tests, σn = 310 kPa, ID0 = 90%, ∆τ = 10 kPa
with rough plate
201
0.0 0.0
ID0 = 30%, Kcmy1/2Kpeak ID0 = 90%, Kcmy1/2Kpeak
-0.2
-0.1
[u]cm (mm)
[u]cm (mm)
-0.4
-0.2
-0.6
20LM60 20DM60
20LM120 -0.3 20DM120
-0.8
20LM310 20DM310
Kcm0,27
20LM200 ; +/- = +/- 0.27 Kcm+/-
20DM200 ; eta = +/-
0,330.33
-1.0 -0.4
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N (-) N (-)
3 1.0
ID0 = 30%, Kcmy1/2Kpeak ID0 = 90%, Kcmy1/2Kpeak
2
0.5
[w]cm (mm)
[w]cm (mm)
1
0 0.0
-1 20LM60
20DM60
20LM120 -0.5
20DM120
-2 20LM310 20DM310
Kcm0,27
20LM200 ; +/- = +/- 0.27 20DM200 ; Kcm = +/- 0.33
-3 -1.0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N (-) N (-)
Figure 3.42: Interpolation of cyclic CNL tests, σn = 200 kPa, ∆τ = 20 kPa with rough
plate: (left) on loose sand (ID0 ≈ 30%), ηcm = ± 0.27 ; (right) on dense sand (ID0 ≈ 90%),
ηcm = ± 0.33
3.3.4 Validation of the identification (CNL) model to the variation of

cyclic amplitude and CNS condition
A simplication of the model for varying the cyclic amplitude and CNS condition is also de-
rived from the analytical formulations of CNL condition. This is verified by the parameters
of the constitutive law which is capable of connecting the cycles of different characteristics,
using finite increments themselves in the parts of complete analytical formulations. The
extension of CNL analytical formulations to the variation of cyclic amplitude is first con-
sidered. Then the modification of CNL analytical formulations by taking into account the
effect of k is applied for CNS condition.
• Procedure for modeling the variation of cyclic amplitude
Once various cyclic amplitudes are applied in succession (∆N ), the corresponding state
parameters for the accumulation rate of displacements are associated. Kaggwa et al. [1991]
proposed the procedure for analyzing effects of irregular cyclic loading (changing the ampli-
tude of cycles in succession) by evaluating the equivalent number (Neq ) of the uniform stress
cycles. The meaning of Neq is the number of cycles at the following loading package with
the corresponding state parameters (e.g. ηi , ∆ηi ) that would cause the same displacements
which result from the previous package of loading with the corresponding state parameters
(e.g. ηi-1 , ∆ηi−1 ). Figure 3.43 shows the evaluation of Neq when various cyclic amplitudes
are applied in succession.
202
[u ]
'K 2
[u] Neq 'K1

[u ] 'N 1
N
N eq 'N1
Figure 3.43: Evaluation of equivalent number of cycles at different cyclic stress conditions
To evaluate the incremental displacemments due to cyclic loading package in succession

(∆Ni ), the procedures are the following;
1. Initialization:
∀ ID0 and/or γd0 , σn0 , ηcm , ∆η0
[u]cm 1 = [u]cm 0 , [w]cm 1 = [w]cm 0 , γd0 = γd1 , ηcm 1 = ηcm 0 , ∆η0 = ∆η0
For i = 1 to IM ; 1st package (∆N1 )

2. Iterative search of Neq at the beginning of step i for:
• [u]cm i corresponding to [u]cm i−1 or γd i−1 for the parameters σn i , ηcm i and ∆ηi
• [w]cm i corresponding to [w]cm i−1 for the parameters σn i , ηcm i and ∆ηi
It is worth noting that Neq for [u]cm i are different from Neq for [w]cm i .
3. Compute increment of ∆[u]cm i and ∆[w]cm i resulting from Neq + ∆Ni with analytical
formulations
4. Update accumulation of displacements at the end of loading package i by:
[u]cm i+1 = [u]cm i + ∆[u]cm i
[w]cm i+1 = [w]cm i + ∆[w]cm i

• Procedure for modeling CNS cyclic interface
In case of CNS condition the procedures are carried out in a different way. Along the CNS
direct shear path, a finite increment of the variables is considered, at an incremental level, as
a combination of a CNL incremental path and of an oedometric incremental path according
to the formal schemes described in Figure 3.44. It can be noted that the memory parameter
is [u]cm .
203
Vn cm W
Vn cm1 CNL; Vn cm1 V cm2

CNL; n
CNL; Vn cm1
CN oedometric; Vn cm1t Vn cm2
Vn cm2 S oedometric; Vn cm1t Vn cm2
CNL; Vn cm2
'N1 'N2 N 'N1 'N2 N
[u] [w]
V cm2 V cm2
CNL; n CNL; n
CNL; Vn cm1 CNL; Vn cm1
CNS oedometric; Vn cm1t Vn cm2 oedometric; Vn cm1t Vn cm2
'N1 'N2 N 'N1 'N2 N
Figure 3.44: Schemes of a combination of a CNL incremental path and of an oedometric

incremental path
To model CNS cyclic interface more satisfactorily, some modifications by considering the
effect of k are applied as follows:
P
• The N interval of interest is discretized into finite incremental Ni steps with Ni =
N , imax = IM
• The effect of k on the interpolation or weight functions (Wc and Wc ) is taken into
account by imposing the multiplicative factor on ηcm2 which can be rewritten as:

∗ −k −k
ηcm2 = ηcm2 exp + 1 − exp ηcm (3.61)
1000 1000
• Due to the variation of σn , the variation of [u]cm has to be interpreted by taking into
account the contribution of passive part and the variation of the sample height. Figure
3.45 shows the interpretation of ∆[u]cm due to the effect of the imposed normal stiffness
(k) and the normal stiffness of sample during shear loading (kn sam ). The relationship
between ∆[u]cm in case of CNL and of CNS can be expressed as:
kn sam
∆[u]cm CNS i = ∆[u]cm CNL i (3.62)
kn sam + k
Using this result 3.62, an expression for the variation in normal stress (∆σn ) is ob-
tained as:
204
Kn sam
'V n CV (if '[u] = 0)
Kn sam
'[u]CNL (if 'Vn = 0)
Figure 3.45: Interpretation of the ∆[u]cm due to the effect of the external imposed normal
stiffness (k) and of proper normal stiffness of the sample (kn sam )
k ∗ kn sam
∆σn cm i = ∆[u]cm CNL i (3.63)
kn sam + k
Figure 3.46 can also clarify the proccedure of the evaluation of the incremental displacements
as well as the variation in normal stress in case of cyclic CNS condition.
As can be seen from the flowchart, this explicit method is easily implemented. However,
the step size (∆Ni ) effect has to be taken into account. When the model contains a very
large step size, the efficiency of explicit integration is compromised severely. To overcome
this deficiency, a smaller step size is used. Figure 3.47 shows the effect of step size for σn0
= 100 kPa, ηcm0 = 0.50, k = 1000 kPa/mm on dense sand. It is evident that the evolution
of σn0 and ηcm0 with a smaller step size (∆Ni = 10) is more steady in the first five hundred
cycles than that with a larger step size (∆Ni = 200).
Eventually, a simple analytical formulation of N at the critical state (Ncrit : the number of
cycles cannot evolve further) is proposed. This can be achieved by taking into account the
influence of ID0 , σn0 , ηcm0 , ∆τ and k. Regarding the results of cyclic tests, Ncrit for σn0 =
100 kPa on both densities are summarized in Figure 3.48. With increasing k and ∆τ , Ncrit
decrease significantly. It is worth noting that ηmax is limited within the range of ηmax <
ηcrit .
For the sake of simplicity, Ncrit at ηcm = 0 will first be formulated as a function of k for
each ∆η as:
for loose sand Ncrit 0 = [20.237 ln(∆η) + 18.049] ln(k) + [−174.63 ln(∆η) − 155.45]
(3.64a)
for dense sand Ncrit 0 = [9650 ln(∆η) + 6624.8] ln(k) + [−83377 ln(∆η) − 57458] (3.64b)
205
For i = 1 to IM
Iterative search of Neq for

[u]cm i and [w]cm i at the be-
i+1
ginning of step i corresponding
to [u]cm i−1 and [w]cm i−1
Calculations of ∆[u]cm i and

∆[w]cm i resulting from Ni with
CNL analytical formulations
Updates at the end of step i

[u]cm i+1 = [u]cm i + ∆[u]cm i
[w]cm i+1 = [w]cm i + ∆[w]cm i
σn cm i+1 = σn cm i + ∆σn cm i
from ηcm i+1 and ∆ηi+1
Figure 3.46: Procedure for evaluating the incremental variation of displacemments and
normal stress in CNS condition
160 0.8
ID0 = 90% ; Vn0 = 100 kPa ; ID0 = 90% ; Vn0 = 100 kPa ;
Kcm0 = 0.50 ; k = 1000 kPa/mm Kcm0 = 0.50 ; k = 1000 kPa/mm
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
'N == 50-200
DN 10 'N = 10-200
DN 10
'N == 200-500
DN 200 'N = 200-500
DN 200
40 0.2
0 0.0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
N (-) N (-)
(a) (b)
Figure 3.47: Step size effect (∆Ni = 10 and ∆Ni = 200) on the results of cyclic CNS test,
σn0 = 100 kPa, ηcm0 = 0.50, k = 1000 kPa/mm on dense sand
206
By taking into account the influence of σn0 , the coefficient derived from experimental data
can simply be given as:
for loose sand Cσn0 = 1.5411 ln(σn0 ) − 6.1637 (3.65a)

for dense sand Cσn0 = 1E − 7 (σn0 )3.4512 (3.65b)
20 14
ID0 | 30 % ; Vn0 = 100 kPa ; CNS ID0 | 90 % ; Vn0 = 100 kPa ; CNS
kk 1000 Dn; 'K

0,1 = 0.10
12
= 1000
15 kk 2000 Dn; 'K
= 2000 0,1 = 0.10 Dn; 'K
= 1000
k 1000 0,1 = 0.10
kk 5000 Dn; 'K
= 5000 0,1 = 0.10 10 k 2000 Dn; 'K
= 2000 0,1 = 0.10
Ncrit [x103] (-)

kk 1000 Dn; 'K
0,2 = 0.20 Dn; 'K
= 5000 0,1 = 0.10
Ncrit (-)
= 1000 k 5000
Dn; 'K
= 5000
kk 5000 0,2 = 0.20 8 k 1000 Dn; 'K
= 1000 0,2 = 0.20
10 k 5000 Dn; 'K
= 5000 0,2 = 0.20
6
4
5
2
0 0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 1
Kcm0 (-) Kcm0 (-)
500 2.5E+05
ID0 | 30 % ; Vn0 = 100 kPa ; CNS ID0 | 90 % ; Vn0 = 100 kPa ; CNS
400 2.0E+05
'K==0,05
Dn 0.05
'K == 0,1
0.05 'K==0,1
Dn 0.10
Dn
300 'K == 0,2 1.5E+05 'K==0,2
Dn 0.20
Ncrit (-)
0.10
Ncrit (-)
Dn
'K =0,05
= 0.20
y_0.05 = -24000Ln(x) + 207000 'K==0,4
Dn 0.40
y_0.05 = -51Ln(x) + 440 Dn
'K = 0.40
Dn=0,4 y_0.10 = -13746Ln(x) + 118823
200 y_0.1 = -19.519Ln(x) + 169.36 1.0E+05 y_0.20 = -7457,1Ln(x) + 64095
y_0.2 = -7.3105Ln(x) + 63.31
y_0.40 = -3800Ln(x) + 32600
100 y_0.4 = -3.1Ln(x) + 27 5.0E+04
0 0.0E+00
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
k (kPa/mm) k (kPa/mm)
Figure 3.48: Summary of Ncrit for σn0 = 100 kPa ; (left) on loose sand (ID0 ≈ 30%) ; (right)
on dense sand (ID0 ≈ 90%)
Finally, the formulation of Ncrit for a given ηcm is expressed as:

Ncrit 0
Ncrit = Cσn0 (ηmax − ηcrit ) + 1 (3.66)
−ηcrit
3.3.5 Checking of the proposed analytical models for validation
The results of model prediction of the tests, in which the sequence and amplitude of two
consecutive packages are changed, are summarized in Figure 3.49 and 3.50. It can be seen
that the model predictions of [w]cm on loose sand starting with whether smaller amplitude or
bigger amplitude are quite overestimated while model predictions of [u]cm are sastisfactory.
On dense sand, particularly [u]cm in case of 10 20DM 120, the intensity of the second package
is slightly underestimated, this is due to ηmax of the second package close to ηcar . However,
the model reproduces, with good accuracy, the rate of the intensity of [u]cm and [w]cm
207
in other cases. The results of model prediction in overview are reasonably close to the
experimental results.
In addition, to demonstrate the applicability of the proposed model in predicting cyclic
loading with various packages, the supplementary cases in which five packages each 2,000
cycles are consecutively applied. Two series of amplitude variations (∆τ = 10, 20, 40,
60 and 80 kPa) are applied on both densities. The level of ηcm is set at 0.34 (in very
contractive range) so as to clarify the size effect of cycles. Figure 3.51 shows the overview
of the supplementary cases.
From the results of model prediction as shown in Figure 3.52, starting with the biggest
package (∆τ = 80 in case of S1) causes the reduction in rate of displacement intensity
of the subsequent packages on both densities. It can be seen that at the final number of
cycles the accumulation of the displacement is much less influenced by the sequence of the
packages of cylic amplitude.
Interestingly, the series starting with the biggest package provides considerable [w]cm on
loose sand. This is due to the size of cyclic loading (ηmax close to ηpeak ). From experimental
point of view, the farthest displacement that the interface could undergo is approximately of
6-7 mm, and this only occurs when testing with low level of initial σn . It seems ambiguous
if this series could be carried out in practice. On dense sand, the dilative behaviour hardly
occurs even though ηmax touches ηcar .
0 1.6
10_20LM120 10_20LM120
model model
-0.2 1.2
[w]cm (mm)
[u]cm (mm)
-0.4 0.8
-0.6 0.4
-0.8 0.0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N (-) N (-)
0 1.6
20_10LM120 20_10LM120
model model
-0.2 1.2
[w]cm (mm)
[u]cm (mm)
-0.4 0.8
-0.6 0.4
-0.8 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N (-) N (-)
kPa, cyclic CNL tests on loose sand (ID0 ≈ 30%), cyclic loading package in succession ∆τ
= 10 and 20 kPa with rough plate
208
0.00 0.00
10_20DM120 20_10DM120
model model
-0.05 -0.05
[u]cm (mm)
[u]cm (mm)
-0.10 -0.10
-0.15 -0.15
-0.20 -0.20
-0.25 -0.25
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N (-) N (-)
0.8 0.8
10_20DM120 20_10DM120
model model
0.6 0.6
[w]cm (mm)
[w]cm (mm)
0.4 0.4
0.2 0.2
0.0 0.0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N (-) N (-)
kPa, cyclic CNL tests on dense sand (ID0 ≈ 90%), cyclic loading package in succession ∆τ
= 10 and 20 kPa with rough plate
In case of CNS condition, the significant feature is the reduction of normal stress as a
function of N . Here, the results of model prediction of the reduction in normal stress
relative to the variation of normal displacement as a result of imposing the stiffness k are
only be presented. Modeling of the mean cyclic shear displacement ([w]cm ) will be presented
in the future. Figure 3.53 typically shows the comparison between experimental data and
model prediction with k = 1000 kPa/mm, σn0 = 100 kPa, ηcm = 0.35 and ∆τ = 10 kPa on
dense sand. The signifiant reduction in normal stress as a function of N is well reproduced
on loose sand in case of k = 1000 kPa/mm, σn0 = 310 kPa, ηcm = 0.30 and ∆τ = 10 kPa
(Figure 3.54). The results of model prediction relative to the reduction in normal stress as
a function of N are close to the experimental results. The proposed model well reproduces
the reduction in normal stress influenced by k, ηcm and ∆τ . However, in case of loose sand
in which the tests can be carried out with a small number of cycles the reduction rate of
normal stress reproduced by the proposed model is quite underestimated especially in the
first five cycles (see Figure 3.55). All features of interface model in CNS condition are shown
in Appendix E.
Figure 3.56 and 3.57 show the model prediction of Ncrit . It can be seen that Ncrit from the
model prediction are in accordance with experimental data. However, at ηcm0 = 0 on both
densities, Ncrit from model prediction would be underestimated. The overall agreement
between model prediction and experimental data is satisfactory.
209
0.8 1.0
S1_ID0 = 30%, Vn = 200 kPa, Kcm=0.34 S1_ID0 = 90%, Vn = 200 kPa, Kcm=0.34
Kpeak Kpeak
0.8
0.6 Kcrit
Kcar
0.6 Kcrit
Kcm (-)
Kcm (-)
0.4 Kcar
0.4
'W = 20 'W = 10
'W = 40 'W = 10
0.2 'W = 20
'W = 60 0.2 'W = 40
'W = 80 'W = 60
'W = 80
0.0 0.0
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
N (-) N (-)
0.8 1.0
S2_ID0 = 30%, Vn = 200 kPa, Kcm=0.34 S2_ID0 = 90%, Vn = 200 kPa, Kcm=0.34
Kpeak Kpeak
0.8
0.6 Kcrit
Kcar
0.6 Kcrit
Kcm (-)
Kcm (-)
0.4 Kcar
0.4
'W = 20 'W = 10
'W = 40 'W = 10
0.2 'W = 20
'W = 60 0.2 'W = 40
'W = 80 'W = 60
'W = 80
0.0 0.0
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000
N (-) N (-)
Figure 3.51: Sequence of cylic loading packages, σn = 200 kPa, cyclic CNL tests with rough
plate ; (left) on loose sand (ID0 ≈ 30%): (right) on dense sand (ID0 ≈ 90%)
0.0 0.00
ID0 = 30%, Vn = 200 kPa, Kcm=0.34 ID0 = 90%, Vn = 200 kPa, Kcm=0.34
-0.1 -0.05
S1 S1
[u]cm (mm)
S2
[u]cm (mm)
S2
-0.2 -0.10
-0.3 -0.15
-0.4 -0.20
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N (-) N (-)
12 0.8
ID0 = 30%, Vn = 200 kPa, Kcm=0.34 ID0 = 90%, Vn = 200 kPa, Kcm=0.34
10
0.6
[w]cm (mm)
8
[w]cm (mm)
S1
S2
6 0.4
S1
4 S2
0.2
2
0 0.0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000
N (-) N (-)
Figure 3.52: Model prediction of cylic loading packages in succession, σn = 200 kPa, ηcm =
0.34, cyclic CNL tests with rough plate ; (left) on loose sand (ID0 ≈ 30%): (right) on dense
sand (ID0 ≈ 90%)
210
160 0.8
10DM100_1000k 10DM100_1000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 4000 8000 12000 16000 0 4000 8000 12000 16000
N (-) N (-)
0.02 160
10DM100_1000k 10DM100_1000k
model model
0 120
[u]cm (mm)
Vn cm (kPa)
-0.02 80
-0.04 40
-0.06 0
0 4000 8000 12000 16000 -0.06 -0.04 -0.02 0 0.02
N (-) [u]cm (mm)
Figure 3.53: Comparison between experimental data and the model prediction, cyclic CNS
test with k = 1000 kPa/mm, σn0 = 100 kPa, ηcm0 = 0.35, ∆τ = 10 kPa, rough plate on
dense sand (ID0 ≈ 90%)
400 0.8
10LM310_1000k 10LM310_1000k
model model
300 0.6
Vn cm (kPa)
K cm (-)
200 0.4
100 0.2
0 0
0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500
N (-) N (-)
400 0.1
10LM310_1000k 10LM310_1000k
model model
300 0
Vn cm (kPa)
[u]cm (mm)
200 -0.1
100 -0.2
0 -0.3
-0.3 -0.2 -0.1 0 0.1 0 500 1000 1500 2000 2500
[u]cm (mm) N (-)
loose sand (ID0 ≈ 30%)
211
160 0.8
10LM100_1000k 10LM100_1000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 5 10 15 20 0 5 10 15 20
N (-) N (-)
0.02 160
10LM100_1000k 10LM100_1000k
model model
0 120
Vn cm (kPa)
[u]cm (mm)
-0.02 80
-0.04 40
-0.06 0
0 5 10 15 20 -0.06 -0.04 -0.02 0 0.02
N (-) [u]cm (mm)
loose sand (ID0 ≈ 30%)
100 100 100

ID0 | 30 % ; Vn0 = 60 kPa ; ID0 | 30 % ; Vn0 = 100 kPa ; ID0 | 30 % ; Vn0 = 100 kPa ;
k = 1000 kPa/mm k = 1000 kPa/mm k = 2000 kPa/mm
expB'K = 0.167; hcm0=0,2
exp_Dh=0,167 expB'K = 0.10; hcm0=0,3
exp_Dh=0,10 expB'K = 0.10; hcm0=0,2
exp_Dh=0,10
model expB'K = 0.20; hcm0=0,3
exp_Dh=0,20 model
Dh=0,1(model)
Dh=0,167(model)
10 10 modelB'K = 0.10
Dh=0,1(model) 10
Ncrit (-)
Ncrit (-)
modelB'K = 0.20
Ncrit (-)
Dh=0,2 (model)
1 1 1
0.1 0 0
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
Kcm0 (-) Kcm0 (-) Kcm0 (-)
100 10000
ID0 | 30 % ; Vn0 = 100 kPa ; ID0 | 30 % ; Vn0 = 310 kPa ;
k = 5000 kPa/mm k = 1000 kPa/mm
1000
expB'K = 0.10; hcm0=0,15
exp_Dh=0,10
10 expB'K = 0.20; hcm0=0,15
exp_Dh=0,20
modelB'K = 0.10
Ncrit (-)
Ncrit (-)
Dh=0,1(model) 100
modelB'K
Dh=0,2 = 0.20
(model) expB'K = 0.032; hcm0=0,3
exp_Dh=0,032
model
Dh=0,032(model)
10
1
0 0.1
0.0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
Kcm0 (-) Kcm0 (-)
Figure 3.56: Comparison between experimental data and the model prediction of Ncrit on
loose sand ID0 = 30%
212
10000 100000 100000
ID0 | 90 % ; Vn0 = 60 kPa ; ID0 | 90 % ; Vn0 = 100 kPa ; ID0 | 90 % ; Vn0 = 100 kPa ;
k = 1000 kPa/mm k = 1000 kPa/mm k = 2000 kPa/mm
10000 10000
1000
expB'K = 0.167
exp_Dh=0,10
model
Dh=0,167(model) 1000 1000
Ncrit (-)
Ncrit (-)
100
Ncrit (-)
expB'K = 0.10
exp_Dh=0,10
expB'K = 0.20
exp_Dh=0,20
100 modelB'K = 0.10
100
Dh=0,1(model) expB'K = 0.10
exp_Dh=0,10
10 modelB'K = 0.20
Dh=0,2 (model) model
Dh=0,1(model)
10 10
1
1 1
0 0 0
0.00 0.20 0.40 0.60 0.80 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
Kcm0 (-)
Kcm0 (-) Kcm0 (-)
10000 1000000
100000
1000
10000
100 ID0 | 90 % ; Vn0 = 310 kPa ;
Ncrit (-)
1000
Ncrit (-)
ID0 | 90 % ; Vn0 = 100 kPa ; k = 1000 kPa/mm
k = 5000 kPa/mm
10 100
expB'K = 0.032
exp_Dh=0,10
expB'K = 0.10
exp_Dh=0,10 model
Dh=0,032(model)
expB'K = 0.20
exp_Dh=0,20 10
1 modelB'K = 0.10
Dh=0,1(model)
modelB'K
Dh=0,2 = 0.20
(model) 1
0 0
0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
Kcm0 (-) Kcm0 (-)
Figure 3.57: Comparison between experimental data and the model prediction of Ncrit on
dense sand ID0 = 90%
3.4 Conclusions
The model presented here is an analytical-proposed model in which only four parameters
for identification (i.e. ID0 , σn , ηcm and ∆η) are derived from experimental data. The single
memory parameter is [u]cm (or γd σn ). Then the extension of the identification (CNL)
model to the variation of cyclic amplitude and CNS is also proposed for the validation
purpose. The proposed model is capable of interpreting the behaviour of cyclic granular soil-
structure in both CNL (analytically) and CNS (by finite analytically increments) conditions.
The predictions of the model have been compared with the experimental data and the
accordance is generally satisfactory. All of the main features of interface behaviour under a
large number of cycles, i.e. contraction, dilation, the critical state of stress that the number
of cycles cannot evolve further and more importantly the reduction in normal stress due to
the accumulation of contraction in case of CNS condition are taken into account.
213
Chapter 4
Finite element modeling
4.1 Introduction
In this chapter, the FE Plaxis 2D (version 8.2) program is applied in a few examples to
model the interaface behaviour. This FE program has been used increasingly for modeling
geomechanical projects with several constitutive models. The well known and simple Mohr-
Coulomb approach taking into account five parameters (E, ν, c, φ and ψ) is chosen to
use. The analysis in this chapter is mainly oriented to highlight the mechanisms existing
within the direct shear test considered as element test, as well as during the load transfer
mobilization of the loaded piles. Preliminary Plaxis implementation of monotonic soil-
structure interface direct shear tests is first presented for providing the applicability of
this software and also focuses the possible defaults of the direct shear tests. Then, an
implementation of some centrifuge model tests on pile carried out at the IFSTTAR in
Nantes (France) within the SOLCYP project is typically presented.
4.2 Direct shear box
This study attempts to illustrate preliminary Plaxis implementation of soil-structure in-

terface direct shear tests (e.g. Arslan [2005]). In this section, a simple implementation
of monotonic CNL and CNS interface direct shear tests is presented. In Plaxis, interfaces
are normally modeled with Mohr-Coulomb model. This model is a perfectly elastoplastic
model involving five input parameters, i.e. E and ν for soil elasticity, ϕ and c for soil
plasticity, and ψ as an angle of dilatancy, and provides a first reliable approximation of
soil behaviour (Brinkgreve et al. [2008]). The Mohr-Coulomb criterion used to desribe the
interface behaviour can be distinguished as:
• For elastic part (with a small displacement), the shear stress τ is given by:
τ < σn tan ϕi + ci (4.1)
214
• For plastic part (with an occurrence of permanent displacement), τ is then given by:
τ = σn tan ϕi + ci (4.2)
where ϕi and ci are the friction angle and cohesion of the interface.
The strength of interface which mainly depends on the roughness surface are relative to the
strength of a soil adjacent to the interface according to the reduction factor for interface
friction (Rinter ) :
ci = Rinter · csoil (4.3a)
tan ϕi = Rinter tan ϕsoil ≤ tan ϕsoil (4.3b)
ψi = 0 ◦ for Rinter < 1, otherwise ψi = ψsoil (4.3c)
4.2.1 Monotonic CNL direct shear box
• Boundary conditions
The interface direct shear is modeled by using plane - strain model with 15-node elements. A
steel shear box containing a sand sample of the sand - structure interface has a dimension of
φ = 60 mm with sample height of 20 mm. Figure 4.1 illustrates the simplified configuration
of interface direct shear test in case of CNL condition. The interfaces are created between
steel shear box-sand, steel palte-sand and steel palte-steel shear box. In this situation the
stress and the prescribed displacement are uniformly distributed on the shear box. The left
hand side of the steel plate is set by vertical fixities so as to allow the upper part (shear
box) displacing horizontally whereas the bottom and the right hand side boudaries are fully
fixed. A medium mesh is sufficient for this simple geometry. The steady pore pressure is
not taken into account because these tests are carried out with dry sand. Six points close
to the interface (i.e. the first three points are positioned above the interface line and the
rest are positioned within the interface zone) are chosen from the midle to the right side of
the shear box for representing the stress - displacement curves.
• Material properties
The material properties used in this study are derived from monotonic CNL direct shear
tests. The interface stiffness is set equal to the elastic soil stiffness. Therefore, the interface
stiffness E = Eur where Eur is proportinal to σn . On the basis of monotonic CNL tests
the model parameters for Mohr-Coulomb model are summarized in Table 4.1. The steel
plate has the same properties as steel shear box. Three values of model parameters on
each density are derived from three levels of initial normal stress σn = 60, 120 and 310
kPa, respectively. The model parameters for steel-steel interface are also defined, which
215
will not affect the model results too much. The influence of surface roughness is able to
be represented by Rinter . To model the interface with very rough plate Rinter = 0.80 is
then be held in this study. It is worth noting that the interface thickness has never been
defined in Plaxis, a virtual thickness factor is then defined. This is a purely numerical
value (0.10 by default). From experimental observations, the deformations and thus the
associated dilatancy commonly take place within the interface shear zone (very thin zone).
To optimise the numerical performance on the interface, the value of virtual thickness factor
= 0.010 is then defined.
A B C
D E F
Figure 4.1: Simplified configuration of an interface direct shear test at CNL condition
Table 4.1: Material parameters for Fontainebleau sand

Parametrs (Unit) Loose sand Dense sand Steel box/plate steel-steel interface
Material model Mohr-Coulomb Mohr-Coulomb Linear elastic Linear elastic
γdry (kN/m2 ) 14.74 16.55 60 -
γsat (kN/m2 ) 20 20 0 -
Eref (kN/m2 ) 1680, 2240, 4260 2010, 4240, 9330 2.1E07 2.1E07
ν 0.30 0.30 0.15 0.15
cref (kN/m2 ) 0.10 0.10 - 0.10
ϕ (◦ ) 30 38 - 5
ψ (◦ ) 0 8 - 0
Rinter 0.80 0.80 - 0.05
• Model procedures for monotonic CNL tests

The value of the applied loads and prescribed displacement can be specified in the input
program. These procedures can be activated or deactivated in the calculation program by
216
means of the staged construction option. The calculation of all phases can be done by means
of the staged construction propcess. All monotonic tests the prescribed shear displacement
is set at 40 mm. The calculation procedures are typically listed in Table 4.2. Figure 4.2
also shows the distribution of vertical displacement due to an application of normal stress
on the sample. It is also found that the distribution of normal stress is less pronounced at
the bottom corners.
Table 4.2: Loading procedures for monotonic CNL test with σn = 120 kPa
Phase Calculation type loading input load (kPa) Displacement (mm)
1 Plastic Staged construction 120 inactive
2 Plastic Staged construction 120 + 40
Figure 4.2: Distribution of vertical displacement due to an application of normal stress σn

= 120 kPa on dense sand
• Analysis of results
The results of monotonic CNL tests are typically shown in Figure 4.3. It can be seen that
the vertical displacement during shearing does not show the significant dilative behaviour
at the interface line. When considering each point selected for representing the stress -
displacement curves, all points also show very small dilation as shown in Figure 4.4. Two
points (”A” and ”B” above the interface line) at the middle would provide more dilation
than the rest but the shear stress is too small. It seems that a simple Plaxis Mohr-Coulomb
model is not fully successful in describing the normal displacement of the interface direct
shear tests. A more advanced model is therefore required. This will be done in the future.
However, in the following figures, as far as the interface strength is concerned, point ”F ”
217
positioned within the interface zone is supposed to be representative for describing the
interface model.
Figure 4.4 then shows the results of monotonic CNL tests in comparison with the experi-
mental data on both densities. It can also be seen that dense sand on which Mohr-Coulomb
model does not include softening behaviour, so after reaching the peak shear stress, the
stress level still evolves constantly. Whereas on loose sand the obtained results from Mohr-
Coulomb model are close to the experimental results.
Figure 4.3: Shear stress and vertical displacement as a result of monotonic CNL condition,
σn = 120 kPa on dense sand
120 0.2
100 0.16
80 0.12
W (kPa)
[u] (mm)
60 0.08
A A
40 B 0.04 B
C C
D D
20 0
E E
F F
0 -0.04
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
(a) (b)
Figure 4.4: Shear stress and vertical displacement at each point of monotonic CNL condi-
tion, σn = 120 kPa on dense sand
218
300 300
ID0 | 30 % ; CNL ID0 | 90 % ; CNL
Vn = 60 kPa
L60_0k Vn = 60 kPa
sn60
250 Vn = 120 kPa
L120_0k 250 Vn = 120 kPa
sn120
Vn = 310 kPa
L310_0k Vn = 310 kPa
sn310
200 plaxis
plaxis 200 plaxis
plaxis
WkPa
7 (kPa)
150 150
100 100
50 50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
(a) (b)
Figure 4.5: Results of monotonic CNL tests: (a) loose sand ; (b) dense sand
4.2.2 Monotonic CNS direct shear box
• Boundary conditions
Imposing a prescribed normal stiffness in Plaxis has never been a direct input. There are
a few ways making this task close to the experimental tests by modifying the the top cap
of the sample as an imposed normal stiffness and loading application. Figure 4.6 shows the
modified configuration of monotonic CNS tests. However, these are few fundamental issues
which would be modified in the future in order to get the reliable results.
Figure 4.6: Modified configuration of an interface direct shear test at CNS condition
219
• Material properties
The material properties used are the same as monotonic CNL direct shear tests. The general
material properties of top cap are the same as steel plate except the elastic stiffness. The
elastic stiffness of top cap (E) modified as an imposed stiffness is expressed as:
∆[u] ∆σ
∆ǫ = = (4.4)
h E
and then the imposed normal stiffness (k) is given as:
∆σ E
k = = (4.5)
∆[u] h
where h is the height of top cap.

• Model procedures for monotonic CNS tests
Unlike CNL tests, the value of the prescribed vertical displacement is specified instead of
the applied loads in the input program. In each case, the prescribed vertical displacement
is derived from the displacement (adjustable according to the elastic stiffness of top cap, E)
due to an applied load from CNL tests (the 1st phase : load application). The calculation
procedures are typically listed in Table 4.3 for σn0 = 100 kPa, k = 1000 kPa/mm on
dense sand. Figure 4.7 shows the distribution of normal stress on the sample due to the
prescribed vertical displacement in case of σn0 = 100 kPa, k = 1000 kPa/mm on dense
sand. It can be seen that at the upper and bottom corners there are respectively more and
less concentrations in normal stress due to the prescribed vertical displacement. During
shearing phase, the prescribed vertical displacement is kept constant so as to describe the
variation of normal stress.
Table 4.3: Loading procedures for monotonic CNS test with k = 1000 kPa/mm , σn0 = 100
kPa on dense sand
Phase Calculation type loading input Displacement (mm)
1 Plastic Staged construction -4.6E-07 (↓) for top cap
2 Plastic Staged construction -4.6E-07 (↓) and +40 (→) for top cap
+40 (→) for shear box
• Analysis of results
Figure 4.8 typically shows the stress paths at each point of monotonic CNS tests for σn0 =
100 kPa, k = 1000 kPa/mm. Again, it can be seen that point ”C” and ”F ” respectively
positioned above and within the interface zone can well represent stress paths according
to the initial conditions. These two points can describe the reduction of normal stress
associated with the shear stress during shearing load on loose sand. Whereas the reduction
of normal stress associated with the shear stress occurs at the beginning of shear loading
and consequently the stress state increases significantly on dense sand.
220
Figure 4.7: Distribution of normal stress due to a prescribed vertical displacement σn0 ≈
100 kPa on dense sand
120 300
ID0 | 90 % ; CNS ;
ID0 | 30 % ; CNS ;
k = 1000 kPa/mm
k = 1000 kPa/mm
Vn0 = 100 kPa
Vn0 = 100 kPa
80 200
A
W (kPa)
W (kPa)
A
B
B
C
C
D
40 D 100
E
E
F
F
0 0
0 40 80 120 160 0 100 200 300 400
Vn (kPa) Vn (kPa)
(a) (b)
Figure 4.8: Stress paths of monotonic CNS tests with k = 1000 kPa/mm , σn0 = 100 kPa
: (a) loose sand ; (b) dense sand
Figure 4.9 also shows the computational results and the experimental data. It can be seen
that the model can reasonably describe the variation of normal stress associated with shear
stress which depends on the initial conditions. However, on dense sand the stress state
continues to gradually increase according to the prescribed shear displacement. To propery
describe monotonic CNS direct shear tests with Plaxis, more advanced model would be
considered.
221
300 400
ID0 | 30 % ; CNS ; ID0 | 90 % ; CNS ;
k = 1000 kPa/mm k = 1000 kPa/mm
Vn0= =6060
sn kPa
kPa Vsnn0==6060kPa
kPa
Vn0= =100
sn 100 kPa
kPa 300 Vsnn0==100
100kPa
kPa
200 V = 310 kPa
snn0= 300 kPa Vsnn0==300
310kPa
kPa
plaxis plaxis
W (kPa)
plaxis plaxis
W (kPa)
Kcrit = 0.566
1 200 Kcrit = 0.566
1
100
100
0 0
0 100 200 300 400 0 100 200 300 400 500
Vn (kPa) Vn (kPa)
300 600
ID0 | 30 % ; CNS ; ID0 | 90 % ; CNS ;
k = 5000 kPa/mm k = 5000 kPa/mm
500 Vn0= =6060 kPa
Vn0
sn = 60= kPa60 kPa
sn kPa
Vn0
sn =kPa100 kPa
= 100
Vn0= =100
sn 100 kPa
kPa
200 Vn0
sn =kPa310 kPa
= 300 400 Vn0= =300
sn 310 kPa
kPa
plaxis
plaxis
»plaxis
W (kPa)
o°¤¼¨
Kcrit = 0.566
W (kPa)
300
1
Kcrit = 0.566
100 200 1
100
0 0
0 100 200 300 400 0 100 200 300 400 500 600 700 800
Vn (kPa) Vn (kPa)
Figure 4.9: Stress paths of monotonic CNS tests with k = 1000 and 5000 kPa/mm , σn0 =
100 kPa : (a) loose sand ; (b) dense sand
4.3 Centrifuge pile tests
In this section some of numerical modeling exercises based on experiments carried out in
the IFSTTAR centrifuge in the SOLCYP project (ANR part) are presented. The physical
modeling test in centrifuge, with a scale of 1/23, is subjected to a centrifugal acceleration of
23∗g. Figure 4.10 shows the details of the experimental setup. The container, equipped with
the instrumentation and the actuator applying the load, is placed in the rotating nacelle of
the centrifuge. An acceleration of 23g is applied at the level of - 240 mm from the surface
of the reconstitued sand sample, corresponding to a radius of 4.943 m with respect to the
axis of the centrifuge. Table 4.4 summarizes the physical prototype of the model and vice
versa, with N = 23.
First, the monotonic tests in tension (pull-out test) is typically modeled with the geometric
scale 1/N as shown in Table 4.5. To model the centrifuge model tests with Plaxis 2D the
boundary conditions are summarized as follows:
• Units: mm, kN, kN/mm, kN/m2 , kN/m3 , Day.
222
It is worth noting that the fictitious time is used to represent the pile head force.
• 2D axisymmetric, y axis of revolution oriented upwards, x axis radial elements 15
nodes.
Vérinéléctrique
electric actuator
Poutresupport
support beam
Capteurs
displacement
sensors
dedéplacement
Pieu
pile
Figure 4.10: Details of the experimental setup of centrifuge pile tests at the IFSTTAR in
Nantes (France) within the SOLCYP project
Table 4.4: Scale factors of centrifuge model tests

Physical parameter Scale factor (scale model/scale prototype)
acceleration N
length 1/N
displacement 1/N
deformation 1
force 1/N 2
mass 1/N 3
strength N
stress 1
weight 1/N 2
density 1
• For the gravity ”1/N” naturally during the growing in ”g”.

• The model pile, aluminum, has a length of 590 mm with an embedding of 560 mm
(above the surface of 30 mm) and a diameter of 18 mm, all leading to a weight of
0.405 kg.
• For the purpose of monotonic calculation, the soil is first divided into three layers:
223
Table 4.5: Summary of centrifuge model tests
Load type Prototype Model
force displacement force displacement
a b a/N 2 b/N
(MN) (mm) (kN) (mm)
Tension 2.12 30 4 1.30
Compression 3.25 28 6.10 1.22
– Top layer; depth 0 to 240 mm

– Middle layer; depth: 240 to 560 mm (pile tip)
– Bottom layer; depth: 560 to 690 mm (under the pile tip)
• The studied area has dimensions height: 720 mm, diameter.
• The pile (radius of 9 mm) is divided into two sections:
– Top section; depth: 0 to 240 mm in the soil
– Bottom section; depth: 240 to 560 mm in the soil
• Two clusters of soil are determined, adjacent to the pile with the thickness of 9
mm, respectively, top and bottom sections for the purpose of cyclic calculation. An
interface is placed against the pile in these two clusters, extended 10 mm below the
tip. The deep layer between 690 and 720 mm is only for the mesh, then it will be
inactivated.
• Mesh is generated by default.
• Pore pressure is zero.
• Weight initialization of soil under 1g: K0 procedure is only activated in the bottom
layer (all other layers inactive).
All details for pile modeling are presented in Figure 4.11. The force acting on the pile head
is linear, placed on the edge of the pile (9 mm in radius). The value of Fy is expressed in
kN/mm, and will be called ”Time” resulting from of using the facility of Plaxis, and then
is given as:
Time = π · 18Fy (4.6)

P
Because ”Time” is proportional to −Mstage .
224
Figure 4.11: Geometry model for pile tests
Physical and mechanical properties of soil and structure elements are determined by using
dimentional analysis in scaling definition. On the basis of centrifuge tests, these parame-
ters of soil model are then determined from the level -240 mm relative to the surface (an
application of acceleration of 23g), corresponding to a radius of 4943 mm with respect to
the axis of the centrifuge.
Soil parameters derived from dimentional analysis in scaling are:
• Specific weight is derived from initial state (dense case) in the prototype: γd = 1704
kg/m3 . For 1g, γd /1g = 1.6716E-08 kN/mm3 . For each layer, γd can be given as:
– Top layer; 0 - 240 mm : γd /1g = 1.6310E-08 kN/mm3
– Middle layer; 240 - 560 mm : γd /1g = 1.7257E-08 kN/mm3
– Bottom layer; 560 - 690 mm : γd /1g = 1.8018E-08 kN/mm3
• Young’s modulus is given according to the depth of each layer: E = 4500 kPa with
reference stress σi = 100 kPa, K0 = n = 0.7, results in E = 5.45 E-05*z + 0.0039 in
kN/mm2 . It is worth noting that z is the opposite of y and taken into account from
y = 690 mm (surface of the container). Figure 4.12 also shows the representation of
Young modulus according to the depth.
• ϕ = 40◦ , ψ = 14◦ , Rinter = 0.97 and c = 1 kPa = 1e-6 kN/mm3
These interface parameters of soil are supposed to model the behaviour of dense sand.
They differ from the parameters measured in Chapter 3, but they are necessary to
225
represent the results of the monotonic centrifuge pull-out pile test.
5.0E-02
4.0E-02
E (kN/mm2)
3.0E-02
2.0E-02
y = 5.45E-05x + 0.0039
1.0E-02
0.0E+00
0 200 400 600 800 1000
Depth (mm)
Figure 4.12: Young modulus (E) according to the depth
• Material parameters for aluminum pile are: E = 67.5 kN/mm2 , γ/1g = 2.487E-08
kN/mm3 and ν = 0.34.
Concerning the interfaces, in Plaxis model interface dilatancy is proportional to the defor-
mations in the interface, in 1/t (t interface thickness). The thickness of interface should be
reduced to increase the deformations therein and thus the associated dilatancy. In mode of
”input”, after preparing the model, the virtual thickness factor is changed to be 0.01 (by
default: 0.1).
4.3.1 Modeling of pile pull-out load test
The pull-out test ”P C74T M ” is selected to typically model with Plaxis. It is the numerical
model corresponding to the typical monotonic test carried out by the IFSTTAR. Point A
is selected on the pile head, on the axis, to track the displacement. The procedures of
calculations are:
• Phase 1: Set up the pile and adjacent soil (except top layer, because the pile emerges
from the surface 30 mm) under 1g. Staged construction, Activation of soil and inter-
faces
P
• Phase 2: Application of 23g. Total multiplier, -Mweight = 23
• Phase 3: Pull-out load application on pile head (4 kN). Staged construction, a reduced-
scale model is given as Fy = 4/π/18 = 0.0707 kN/mm.
It can be noted that the stage of an application of 1g only optimizes the stability of
the rearrangement sand grains subjected to centrifugal acceleration. At pull-out load
application, it displays 4 for ”Time” interval for Fy .
The results of centrifuge model test are summarized as follows:
226
• Application of 1g for the pile and adjacent soil as shown in Figure 4.13:
– Maximum settlement of soil: -0.11 mm (very low because of high modulus), and
pile : -0.052 mm
– At interface; maximum σn = 21.06E-6 kN/mm2 in comparison with γz: 1.24E-05
kN/mm2 and maximum τ = 18.06E-6 kN/mm2 in comparison with K0 γz tan ϕ
: 7.3E-6 kN/mm2
– Saturated friction, slight dilatancy and considerable negative friction
Figure 4.13: Phase1: Application of 1g for the pile and adjacent soil
• Application of 23g for all soil layers and the pile (Figure 4.14):
– Maximum settlement of soil: -2.52 mm and pile : -1.16 mm
– At interface; maximum σn = 362.84E-6 kN/mm2 , in fact 2/3 considered, in com-
parison with 23γz: 2.86E-04 kN/mm2 and maximum τ = 294.82E-6 kN/mm2 ,
in fact 2/3 considered, in comparison with 23K0 γz tan ϕ : 1.68E-4 kN/mm2
– Considerable negative friction and dilatancy
• Pull-out interface friction: the negative skin friction has disappeared, the normal and
shear stresses are of the same sign. Summary verification on the distribution of τ : the
average shear stress τavg ≈ 131 kPa or 1.31E-4 kN/mm2 , and τavg · π · 18 · 560 = 4.15
kN. This is in accordance with the centrifuge model test.
After settling in 23g, the pull-out loading of 4 kN is applied. In comparison with the
centrifuge model test, Plaxis calculation provides a good agreement in the range of
227
pull-out loading of 4 kN as shown in Figure 4.15. However, a defect still exists as a
result of a trend of a continuous increase of load-displacement.
Figure 4.16 shows the distribution of σn and τ along the pile under several loadings. The
load acting on the pile tip as a function of loading on pile head is also presented in Figure
4.17. Since the application of loading in 23g is applied, the load acting on the pile tip
resulting from pull-out loading of 4 kN is less pronounced.
Figure 4.14: Phase2: Application of 23g for the pile and adjacent soil
-5
-4
Pull-out load (kN)
-3
Centrifuge model test
Plaxis calculation
-2
-1
0
0 -1 -2 -3 -4
Displacement (mm)
Figure 4.15: Comparison between centrifuge pile test and Plaxis calculation of pull-out load
4 kN on pile head for P C74T M
228
2.E-04 4.E-04
PC74TM PC74TM
0.E+00 2.E-04
Vn (kN/mm2)
W (kN/mm2)
-2.E-04 0.E+00
1g
23g
1g
pull-out_4 kN
-4.E-04 -2.E-04 23g
pull-out_4 kN
-6.E-04 -4.E-04
0 200 400 600 800 0 200 400 600 800
Height from bottom level (mm) Height from bottom level (mm)
(a) (b)
Figure 4.16: Distribution of σn and τ acting along the pile shaft for P C74T M
4
PC74TM
3
Load_tip (kN)
1g
2 23g
pull-out_4 kN
1
-1
0 1 2 3 4 5
Load_head (kN)
Figure 4.17: Load acting on pile tip as a function of loading on pile head for P C74T M
4.3.2 Modeling of pile pull-out load test (1st cycle)
For cyclic model ”PC84TC” is selected, neither properties of the pile nor mechanical and
physical properties of the soil have changed. It refers to the cyclic pile test C075T05 carried
out by the IFSTTAR. The first cycle of centrifuge test ”PC74TM” is intended to model
with the following extreme forces in tension (Table 4.6):
Table 4.6: Scale factors of centrifuge cyclic model test

PC84TC centrifuge test prototype model
a (MN) 1/23 (kN)
Max tension force on pile head 1.587 a/232 = 3
Min tension force on pile head 0.529 a/232 = 1
229
To model pile under cyclic axial load, the geometry in the initial phase has been changed
(Figure 4.18): 5 layers adjacent to the pile instead of 2 layers, so as to better distinguish
the characteristics of cycles in terms of the depth.
Figure 4.18: Geometry model for cyclic axial load test
• Calculation of loading in 1st cycle

At this stage, ”Time” is still used to designate the force on pile head (kN), although
”Time” is a gradual function. And the linear force in kN/mm will always be assigned
manually under staged construction. From monotonic pull-out test, it will be neces-
sary to ”decode” Time in order that this variable represents the force on pile head
even when it decreases during a cycle. Centrifuge modeling of pile under cyclic axial
load is given in the following procedures:
– Phase 1: Staged construction, Activation of sand and interfaces (1g), except at
30 mm above the adjacent pile
P
– Phase 2: Total multiplier. -Mweight = 23
– Phase 3: Staged construction, Activation of the linear force (0.0177PkN/mm),
equivalent to a loading at pile head up to 1 kN. Time interval = 1 ( Time =
1)
2)
3)
230
– Phase 6: Staged construction, Activation of the linear force (0.0354 kN/mm),
P
equivalent to a loading at pile head up to 2 kN. Time interval= 1 (where Time
= 4, although total force = 2 kN, but this will then be decoded)
– Phase 7: Staged construction, Activation of the linear force (0.0177 kN/mm),
P
equivalent to a loading at pile head up to 1 kN. Time interval = 1 (where
Time = 5, although total force = 1 kN, but this will then be decoded)
From calculation, the ”Time” (Fy by using the facility of Plaxis) in the last two stages
can be modified in order that Time increments higher than 3 are taken negatively,
then afterwards led to 1 kN (Figure 4.19).
5
Pull-out load (kN)
2
PC74TM
1 PC84TC_calculation
PC84TC_modification
0
-1.6 -1.2 -0.8 -0.4 0.0 0.4
Displacement (mm)
Figure 4.19: Initial pull-out load for cyclic axial load P C84T C
2.E-04 4.E-04
PC84TC PC84TC
1.E-19 2.E-04
Vn (kN/mm2)
W (kN/mm2)
-2.E-04 3.E-19
23g
23g
load 1 kN
load 1 kN
-4.E-04 load 3 kN -2.E-04 load 3 kN
unload 1 kN
unload 1 kN
reload 2 kN
reload 2 kN
-6.E-04 -4.E-04
0 200 400 600 800 0 200 400 600 800
(a) (b)
Figure 4.20: Distribution of σn and τ along the pile shaft P C84T C for Fy = 1 → 3 → 1 →
2 kN
231
Figure 4.20 shows the distributions of the normal and shear stresses along the pile during
loading up to 3 kN, unloading to 1 kN and then reloading to 2 kN. The normal stress
increases significantly when the load acting on pile head varies between 1 and 3 kN due to
a trend of dilation. Regarding the shear stress, the negative friction due to the application
of 23g still exists under the loading of 1 kN and then disappears during the subsequent
loading.
4.3.3 Modeling of pile pull-out load test (for further cycles)
Considering the simultaneous variation of σn and τ along the pile during the first cycle,
and the pre-existence of a significant negative skin friction due to the application of 23g. It
seems useful to examine the cyclic path at various depths along the pile, in order to decide
the parameters σn cm0 , τcm0 , ∆τ , and ηcm0 to be taken into account as a function of depth.
Figure 4.21 also shows the stress paths during the first cycle at sand-pile interface.
1.5E-04
1
PC84TC
layer 1_23g RinterǜWDQM
5.0E-05 layer 2_23g
layer 3_23g
W (kN/mm2)
layer 4_23g
layer 5_23g
-5.0E-05 layer 1_N = 1
layer 2_N = 1
layer 3_N = 1
layer 4_N = 1
-1.5E-04 layer 5_N = 1
RinterǜWDQM
1
-2.5E-04
-4.E-04 -3.E-04 -2.E-04 -1.E-04 0.E+00
Vn (kN/mm2)
Figure 4.21: Stress paths along the pile shaft during the first cycle for each layer
The reduction of normal stress under CNS condition can be calculated, for each depth, for
a given number of cycles, wherein k is the imposed normal stiffness expressed as:
2G
k= (4.7)
R
where;
R = 9 mm and G is a linear function with depth as E and G = E/2/(1 + ν)
E = 0.0039 + 5.45E-5*z (kN/mm2 ) and z = 690 - y, height from the bottom level (mm)
The distribution of normal stiffness is also presented in Table 4.7
232
Table 4.7: Distribution of the normal stiffness (k) according to the depth in case of
”PC84TC”
Layer σn cm0 τcm0 ∆τ ηcm0 E G k
2 2
(mm) (kPa) (kPa) (kPa) (-) (kN/mm ) (kN/mm ) (kPa/mm)
1 38.5 19.3 18.0 0.50 6.571E-03 2.527E-03 5.616E+02
2 95.2 53.7 32.5 0.56 1.191E-02 4.581E-03 1.018E+03
3 141 78.1 48.4 0.55 1.785E-02 6.866E-03 1.526E+03
4 153 77.3 69.3 0.51 3.333E-02 1.282E-02 2.849E+03
5 172 28.5 104.2 0.17 3.115E-02 1.198E-02 2.662E+03
From Table 4.7, it can be found that the values of maximum stress ratio (ηmax ) and the
cyclic amplitued ∆η are higher in the upper part of these five layers. This indcates that
these cycles are largely in contracting domain. It is interesting to consider the following
cycles which show that the former is largely influenced by the initial negative skin friction,
while the following cycles are ”stabilized”.
The pseudo-creep describing the cyclic evolution starts just after the application of the mean
cyclic load (2 kN) at the pile head. According to the parameters shown in Table 4.7 and to
the formulation of Ncrit developed in Chapter 3 for dense sand, the local evolution of initial
stress state (σn cm0 , τcm0 and ηcm0 ) to critical (σn cm crit , τcm crit and ηcm crit ) condition can
be calculated for each layer. Figure 4.22 shows the evolution of normal stress as a function
of number of cycles. It can be seen that the reduction of normal stress significantly increases
during the first 2,000 cycles and then the reduction rate is almost constant when N > 2000.
200
PC84TC
layer_1 (top)
150 layer_2
layer_3
Vn (kPa)
layer_4
layer_5 (bottom)
100
50
0
0 2000 4000 6000 8000 10000 12000
N (-)
Figure 4.22: Evolution of σn along the pile shaft P C84T C as a function of N
The more difficult task is to apply this change in normal stress along the pile shaft. For this
purpose, a concentric soil layer of thickness equal to the pile radius has been defined for each
horizontal layer. Then the facility developed in Plaxis is used to prescribe a volume change
in a predefined cluster. The distribution of normal stress change is obtained by applying
a distribution of volume change in these clusters. The definition of the distribution of
233
volume change is made by iterations as a volume change in a cluster has an influence in
the neighbor clusters. In this case the distribution of normal stress acting on the pile shaft
which is calculated after 18 iterations is presented in Table 4.8.
Table 4.8: Distribution of the normal stress calculated from the application of volume strain
in predefined cluster adjacent to pile shaft in case of ”PC84TC”
Layer 1 2 3 4 5
imposed ǫv -3.31 -3.61 -4.08 -4.13 -6.39
σn 28.8 75.6 105.9 114.1 71,3
% σn 65.5 67.1 78.0 67.13 73.5
Figure 4.23 shows the cyclic pull-out load - displacement curve at pile head for 2,000 cycles
of centrifuge model test in comparison with Plaxis calculation. In the phase where the load
is constant (2kN), it is the mean cyclic load (pseudo-creep). The result of cyclic pseudo-
creep from Plaxis calculation is close to the centrifuge model test. The distribution of
normal and shear stresses acting on the pile shaft after the application of 2,000 cycles is
also presented in Figure 4.24. It is found that under cyclic pull-out loading the stress state
at pile tip is less prominent.
-4
-3
Pull-out load (kN)
-2
Plaxis calculation
-1
0
0 -0.5 -1 -1.5 -2 -2.5
Displacement (mm)
Figure 4.23: Comparison between centrifuge pile test and Plaxis calculation of cyclic pull-
out load 2 kN for 2,000 cycles of P C84T C
234
1.E-04 4.E-04
PC84TC PC84TC
0.E+00
2.E-04
Vn (kN/mm2)
-1.E-04
W (kN/mm2)
-2.E-04 3.E-19
-3.E-04
23g -2.E-04
23g
-4.E-04 N=1 N=1
cyclic pseudo-creep: N = 4000 cyclic pseudo-creep: N = 4000
-5.E-04 -4.E-04
0 200 400 600 800 0 200 400 600 800
(a) (b)
Figure 4.24: Distribution of normal and shear stresses acting on the pile shaft ”PC84TC”
4.3.4 Modeling of pile compression load test
The monotonic compression loading of ”PC74CM” is modeled in the same way as mentioned
in pull-out test. After the application of 23g, the compression load is then applied with Fy =
6.1/π/18 = 0.1078 kN/mm. Figure 4.25 shows the distribution of normal and shear stresses
after applying the compression loading 6.1 kN. The negative skin friction disappears after
applying the compression loading. The normal and shear stresses increase as a function of
depth and these are considerable at pile tip in comparison with pull-out test (Figure 4.26).
2.E-04 2.E-04
PC74CM PC74CM
0.E+00 0.E+00
Vn (kN/mm2)
W (kN/mm2)
-2.E-04 -2.E-04
1g
-4.E-04 -4.E-04 23g
compression_6.1 kN
1g
-6.E-04 23g -6.E-04
compression_6.1 kN
-8.E-04 -8.E-04
0 200 400 600 800 0 200 400 600 800
(a) (b)
Figure 4.25: Distribution of σn and τ acting along the pile shaft for P C74CM
Again, the summary verification on the distribution of τ shows that the average shear stress
τavg ≈ 184 kPa or 1.84E-4 kN/mm2 , and τavg ·π ·18·560 = 5.829 kN. The agreement between
Plaxis calculation and centrifuge model test is satisfactory.
235
Considering the displacement of pile head, after settling in 23g, the result from Plaxis
calculation is in accordance with centrifuge model test in the range of compression loading
of 6.1 kN. A trend of continuous increase of load-displacement from Plaxis calculation still
exists.
Figure 4.28 also shows the load acting on the pile tip as a function of loading on pile head.
Since the application of loading in 23g is applied, the load acting on the pile tip resulting
from compression loading of 6.1 kN is more prominent.
2.E-04 6.E-04
4.E-04
0.E+00
Vn (kN/mm2)
2.E-04
W (kN/mm2)
-2.E-04
0.E+00
-4.E-04
-2.E-04
-6.E-04
monotonic pull-out_4 kN -4.E-04
monotonic pull-out_4 kN
monotonic compression_6.1 kN monotonic compression_6.1 kN
-8.E-04 -6.E-04
0 200 400 600 800 0 200 400 600 800
(a) (b)
Figure 4.26: Comparison of distribution of σn and τ acting along the pile shaft for pull-out
(P C74T M ) and compression (P C74CM ) tests
8
Compression load (kN)
4
Plaxis calculation
2
0
0 1 2 3 4 5
Displacement (mm)
Figure 4.27: Comparison between centrifuge pile test and Plaxis calculation of compression
load 6.1 kN
236
2.0
PC74CM
1.6
Load_tip (kN)
1.2
0.8
1g
0.4 23 g
compression_6.1 kN
0.0
0 2 4 6 8
Load_head (kN)
Figure 4.28: Load acting on pile tip as a function of loading on pile head for P C74CM
4.3.5 Modeling of pile compression load test (1st cycle)
The procedure of modeling cyclic compression test is similar to that as described in cyclic
pull-out test. Again, the first cycle of centrifuge test ”PC74CC” is selected to typically
model cyclic compression test. Table 4.9 shows the extreme forces in cyclic compression
test.
Table 4.9: Scale factors of centrifuge cyclic compression test

PC74CC centrifuge test prototype model
a (MN) 1/23 (kN)
Max compression force on pile head 1.904 a/232 = 3.6
Min compression force on pile head 0.635 a/232 = 1.2
Following the procedures for the first cycle as mentioned in cylic pull-out loading, the
”Time” (Fy ) is modified in order that Time increments higher than 3.6 are taken negatively,
then afterwards led to 1.2 kN (Figure 4.29). Figure 4.30 shows the stress distribution along
the shaft pile during the first cycle. Similarly to cyclic pull-out test, during the first cycle
in compression the negative skin friction along the pile shaft disappear as a result of an
increase of loading step.
237
5
PC74CC
Load (kN)
3
N=1
2
0
-2.5 -2.0 -1.5 -1.0 -0.5 0.0
Displacement (mm)
Figure 4.29: Initial compression load for cyclic axial load P C74CC
1.E-04 2.E-04
PC74CC PC74CC
0.E+00 1.E-04
W (kN/mm2)
-1.E-04 0.E+00
Vn (kN/mm2)
-2.E-04 -1.E-04
23 g 23 g
-3.E-04 load 1.2 kN -2.E-04 load 1.2 kN
load 3.6 kN load 3.6 kN
-4.E-04 unload 1.2 kN -3.E-04 unload 1.2 kN
reload 2.4 kN reload 2.4 kN
-5.E-04 -4.E-04
0 200 400 600 800 0 200 400 600 800
(a) (b)
Figure 4.30: Distribution of σn and τ along the pile shaft P C74CC for Fy = 1.2 → 3.6 →
1.2 → 2.4 kN
4.3.6 Modeling of pile compression load test (further cycles)
During the first cycle in compression, the stress state acting along the pile shaft varies.
Figure 4.31 shows the stress paths during the first cycle. Similarly to cyclic pull-out load
test, since 23g is applied, the stress state for each layer is close to the limit state. After
the application of the first cycle, the stress state for each layer returns into the contractive
domain. The pseudo-creep describing the cyclic evolution then starts from the mean cyclic
load (2.4 kN) and the initial state can be given as shown in Table 4.10. It can be noted
that τcm0 and ηcm0 are derived from the iterations.
The evolution of normal stress during cyclic loading until reaching the critical state is also
derived from the iterations of volume strain application in predefined cluster adjacent to
238
the pile shaft. Table 4.11 shows the imposed volume strain in predefined cluster adjacent
of pile shaft resulting in the variation of normal stress.
1.5E-04
1
PC74CC
layer 1_23g RinterǜWDQM
5.0E-05 layer 2_23g
layer 3_23g
W (kN/mm2)
layer 4_23g
-5.0E-05 layer 5_23g
layer 1_N = 1
layer 2_N = 1
layer 3_N = 1
-1.5E-04 layer 4_N = 1
RinterǜWDQM
layer 5_N = 1
1
-2.5E-04
-4.E-04 -3.E-04 -2.E-04 -1.E-04 0.E+00
Vn (kN/mm2)
Figure 4.31: Stress paths along the pile shaft during the first cycle in compression for each
layer
Table 4.10: Initial state of evaluating the cyclic evolution according to the depth in case of
”PC74CC”
Layer σn cm0 τcm0 ηcm0 ∆η ηmax
(mm) (kPa) (kPa) (-) (-) (-)
1 35.7 19.8 0.515 0.410 0.805
2 87.9 12.0 0.132 0.429 0.357
3 113.7 37.5 0.316 0.452 0.583
4 122.7 46.0 0.515 0.415 0.707
5 263.7 135.9 0.520 0.429 0.732
Table 4.11: Distribution of the normal stress calculated from the application of volume
strain in predefined cluster adjacent to pile shaft in case of ”PC74CC”
Layer 1 2 3 4 5
imposed ǫv -2.45 -2.18 -2.17 -2.54 -2.61
σn 16.2 40.0 54.2 44.7 142.7
σn crit 22.6 14.3 44.1 77.8 168.3
Figure 4.32 shows the cyclic compression load - displacement curve at pile head for 2,000
cycles. The mean cyclic displacement at pile head begins from the mean cyclic load 2.4
kN. It can be seen that the displacement during the pseudo-creep phase in case of cyclic
compression is less pronounced in comparison with cyclic pull-out load test. The difference
between the numerical modeling and the centrifuge test is important in this case (compres-
239
sion). It is probably due to the point effect not suitably simulated at large stressess in
Plaxis with Mohr-Coulomb model.
Figure 4.33 also shows the distribution of normal and shear stresses acting on the pile shaft
after the application of 2,000 cycles. The reduction in normal stress increases as a function
of depth after applying 2,000 cycles. The normal stress is even small in comparison with the
application of 23g. The negative skin friction along the pile shaft disappears after applying
2,000 cycles and the skin friction along the pile shaft slighly decreases resulting from the
reduction in normal stress.
4
Compression load (kN)
2
Plaxis calculation
1
0
0 0.5 1 1.5 2 2.5
Displacement (mm)
Figure 4.32: Comparison between centrifuge pile test and Plaxis calculation of cyclic com-
pression load 2.4 kN for 2,000 cycles of P C74CC
1.E-04 2.E-04
PC74CC PC74CC
0.E+00
1.E-19
Vn (kN/mm2)
W (kN/mm2)
-1.E-04
-2.E-04 -2.E-04
-3.E-04
23g -4.E-04
23g
-4.E-04 N=1 N=1
cyclic pseudo-creep: N = 4000 cyclic pseudo-creep: N = 4000
-5.E-04 -6.E-04
0 200 400 600 800 0 200 400 600 800
(a) (b)
Figure 4.33: Distribution of normal and shear stresses acting on the pile shaft ”PC74CC”
240
4.4 Conclusion
FEM Program Plaxis, with the simplest constitutive model: Mohr-Coulomb with 5 param-
eters is used to model the interface behaviour. The first analysis devotes to model the
behaviour of interface direct shear box. Since the model is 2D in plane strain, and not 3-D,
the defaults (inhomogeneity) of the direct shear test are clarified, perhaps magnified. It is
found that a simple Plaxis Mohr-Coulomb model is not fully successful in describing the
normal displacement of the interface direct shear tests. As regards the interface strength,
the obtained results from Mohr-Coulomb model are close to the experimental results.
The second part of our FEM simulations is related to pile tests. This deals with centrifuge
model pile tests performed at the IFSTTAR in Nantes, France. Pull-out pile tests and
compression pile tests, under monotonic and cyclic loading, are intended to model in order
to develop an alternate methodology compared to the existing ones. In this part, the
methodology uses the FEM program as a tool for managing the soil, the pile, and their
equilibrium. The interface cyclic pseudo-creep developed in the framework of CNS curves
function of the number of cycles is introduced. The complete approach would require the
integrated curves along CNS paths. A simplified analysis, used in these computations,
refers to the critical number of cycles (Ncrit ) previously developed and to a supposed law of
evolution. Another simplification of this computation is to apply the drop of normal stress
in one phase only, since the cyclic parameters evolve during this drop. This methodology
is developed, but many refinements remain to be done.
241
Chapter 5
Calculation of real piles subjected

to cyclic loading, for civil
engineering projects
5.1 Introduction
Many efforts have been devoted for calculating real piles since many years: prediction of
load transfer, head stiffness (relation load-displacement), stability/safety of piles axially
loaded in the framework of civil engineering projects. The t-z method remains the favourite
one of the most part of designers due to its simplicity. Another working method is the finite
element or finite difference method, using suitable constitutive equations. The heaviness
of the approach and the duration of computations represent a real obstacle against this
kind of analysis. Both kinds of approaches need reliable soil and interface parameters. In
addition modeling the real mechanisms of mobilisation of shaft and point resistance seems
to be a fruitful working way. The work developed in this thesis would propose a compromise
between these two extreme points of view. Nevertheless a lot of improvements of our analysis
remain necessary.
5.2 Behaviour of real piles subjected to axial cyclic loading
In case of axially loaded piles, only considering the geometrical effect resulting in the in-
tensity of stress transfer obviously inverse function of the current distance to the pile axis
leads to give priority to the soil-structure interface. The interface is the locus of this load
transfer mechanism by friction. In the radial direction, on one side, the pile is a nearly
incompressible material. On the other side, the soil prescribes constraints which are in
the radial direction not far from a pressuremetric one, and in the axial direction a shear
stress-tangential relative displacement relation. These two constraints should be compatible
242
with a third constraint: the friction. From this step it is clear that interface shear paths at
prescribed normal stiffness (CNS) should govern the load transfer.
The simple model of lateral friction developed by Boulon & Foray [1986] - Figure 1.16,
formula (1.8)- gives an evaluation of the normal stiffness function of the pressuremetric
modulus of the soil (Ep ) and the radius of pile (R). For bored or cast in-situ piles, the
pressuremetric modulus could be considered in a small deformation and is not far from
the shear modulus (G). But for driven piles, large radial deformations are prominent and
it is necessary to consider the real pressuremetric modulus (large strains). The range of
parameters of CNS direct shear tests relevant to piles are at least the initial normal stress
and the normal stiffness. The range of these parameters will be presented and discussed in
the next section.
The finite element or finite difference method should be considered as a tool for modeling
the volume load transfer from small distance to the pile axis (often at the limit strength) to
the large ones (far from the soil strength), insuring in addition a precise local equilibrium
(the minimum requirement of mechanics). Now, what about the constitutive equations?
Modeling cyclic loading needs thinking at the cyclic mean paths when considering a large
number of cycles, excepted if the confidence ratio on incremental constitutive equations
reaches 100 %! For reducing computation time, the number of cycles (series of cycles)
instead of time (within a single cycle) for integrating adequate constitutive equations and
reducing computation time will be treated. For soil-structure interface, the related tests
and models should describe the so-called ”cyclic pseudo-creep” where the a progression of
negative drop in normal stress occurs.
5.3 Determination of soil and interface parameters
The ordinary data of soil such as the distribution of weight density, water content, strength
(c, ϕ, ψ) and stiffness (G for small strains, Ep for large strains) are necessary for the calcu-
lation of pile foundation. If the properties are known only at some depths, this distribution
has to be interpolated or extrapolated. In case of G and Ep , using the power law of Kondner
would be useful, assuming a classical K0 (≈ 0.50) for bored piles and a high value (≈ 1 or 2)
for driven piles. A linear approximation of soil modulus versus depth (see Young modulus
(E) according to depth in previous chapter) represents a single and suitable choice.
Concerning the soil-pile interface, a small number of cyclic CNS tests according to the
distribution of the main cyclic parameters along the pile (effective normal stress depending
on the installation mode, mean initial shear stress, cyclic amplitude and normal stiffness)
would be sufficient. These tests are supposed to indicate the negative drop in normal stress
as a function of the number of cycles.
243
5.4 Model prediction of piles under cyclic axial loading
5.4.1 Initial state (pile installation)
This state is the result of the installation mode. For bored or cast in-situ piles, after the
calculation of geostatic state (K0 ), the domain of soil corresponding to the place of pile is
generally replaced by the material of pile. If the pile is made of concrete the initial normal
stress could slightly overpass the K0 stresses because the concrete is initially liquid. For
driven piles, the installation generates a large positive drop in normal stress as well as a
strong negative skin friction. In these two cases (bored and driven methods), the operator
has to find a strategy in order to take into account these features. For the normal stress, a
positive volume change around the pile could be a fair solution. On the other hand, there is
no very satisfactory solution for modeling the negative skin friction and its consequence in
the surrounding soil. A simple way would artificially be to increase or decrease the weight
density of the pile.
5.4.2 Characterizing the loading acting on pile head and along pile shaft
It is necessary to analyse and to simplify the probable head loading program of the pile,
then the most severe sequences of cyclic loading would be selected. Only these extreme
sequences which could drastically modify the long term pile capacity are considered.
The interface phenomena are privileged as the amplitude
H of the cycles within the soil itself
rapidly decrease with the distance from the pile (see 5.2). A formulation of cyclic pseudo-
creep for modeling a large series of cycles is considered. If the formulations of volumetric
pseudo-creep are available (e.g. Wichtmann [2005] and Wichtmann et al. [2005]), these
formulations would also be used for the surrounding soil at large distance from the pile.
However, it can be noted that the influence of surrounding soil at large distance from the
pile remains relatively small.
A first preliminary FEM calculation seems necessary for roughly defining the local cycles
along the pile shaft, due to the extreme loading previously defined. This calculation should
then give the local cyclic parameters (mean cyclic normal stress, mean cyclic shear stress
and cyclic amplitude).
H The normal stiffness which is a very important parameter still has
to be investigated ( 5.3).
5.4.3 Soil and soil-structure interface tests
.
From ordinary soil experiments whether under monotonic or cyclic loading, the informa-
tions (e.g. weight density, strength, stiffness) are obtained. Concerning the soil-structure
H
interface, the informations resulting from the preliminary FEM calculation ( 5.4.2) are
useful for precisely defining a limited number of CNS tests supposed to properly cover the
244
range of local cyclic parameters. The main result of these tests is the negative drop in
normal stress at some selected depths along the pile.
A simple interpolation is necessary for generating this drop at any level along the pile shaft
by generalizing the results of the cyclic CNS tests. Before explaining the phases of the final
FEM calculattion, it is useful to consider the way of using these cyclic CNS interface tests
in full details. Figure 5.1 shows the typical cyclic CNS paths in stress plane. The cyclic
CNS path starts from the initial stress state (σn cm0 , τcm ):
τcm
ηcm0 =
σn cm0
W (kPa)
Kcrit
1
Kcm0
Wcm 'W CNS (k) 1
Vn cm0 Vn (kPa)
Figure 5.1: Typical cyclic CNS paths of soil-structure interface
The behaviour of the interface is always contractive as can be seen in the experimental
section. Then, the center of cycles will progressively move to the ”critical state” (σn cm crit ,
τcm )
τcm
σn cm crit = (5.1)
ηcrit
It can be noted that the top of the cycles reaches the critical line before the center of cycles.
For this reason, the rupture of the interface could overcome ealier, for σn cm crit1 :
τcm + α ∆τ
2
σn cm crit1 = (5.2)
ηcrit
where: 0 < α < 1.

√
From the experimental point of view in this study, the coefficient α is not far from 22 which
represents the ”reduced efficient value” of the cyclic amplitude. Therefore, the maximum
negative drop in normal stress ranges between ∆σn and ∆σn1 :
245

1 1
∆σn = σn cm crit − σn cm0 = τcm − (5.3)
ηcrit ηcm0

τcm ∆τ τcm
∆σn1 = σn cm crit1 − σn cm0 = 1+α − (5.4)
ηcrit 2τcm ηcm0
or it can be rewritten as:
∆σn ηcm0
= −1 (5.5)
σn cm0 ηcrit
∆σn1 ηcm0 ηcm0 ∆τ

= −1+α · (5.6)
σn cm0 ηcrit ηcrit 2τcm
Let us recall the whole set of 8 parameters governing the evolution of normal stress during
a cyclic CNS test: roughness surface, ID0 , σn cm0 , τcm , ∆τ , ∆σn , N and k. For a given
contact, density of soil, normal stiffness and according to Vashy-Buckingham theorem, these
parameters can be reduced to 4 parameters:
∆σn ∆τ
, , ηcm0 , N
σn cm0 τcm
Figure 5.2 gives an example of the presentation of the results of cyclic CNS tests directly
related to the evolution of local normal stress along pile shaft under axial cyclic loading.
This figure shows the reduced drop in normal stress as a function of the reduced cyclic shear
stress and of the number of cycles. It can be seen that the lower limits range between σ∆σ n
n cm0
and σ∆σ n1
n cm0
. These results have to be generalized to the other values of ηcm0 and k.
0.1
0
N = 50
-0.1
'Vn/Vn cm0
N = 500
N = 1000
-0.2
-0.3
-0.4
-0.5
0.0 0.2 0.4 0.6
'WWcm
Figure 5.2: Typical results of CNS soil-structure interface shear tests directly related to the
change in normal stress along pile shaft under axial cyclic loading; ID0 = 90 %, ηcm0 = 0.35
, k = 1000 kPa/mm
246
5.4.4 FEM modeling the behaviour of a pile under axial cyclic loading
Having prepared this model, the successive phases of this process can be described in the
following:
• Phase 1: Initialisation of the geostatic stresses of the FEM model.
• Phase 2: FEM modeling the pile installation.
• Phase 3: For each series of N cycles of the global loading, step to step, FEM calculation
of 1 or 2 cycles of loading.
• Phase 4: Extraction of the local cyclic parameters along the pile shaft. Rough solution:
at some selected points along the pile shaft (5 to 10). More refined solution: at each
integration point along the pile shaft.
• Phase 5: Calculation of the local corresponding negative drop in normal stress by
using the diagrams as typically shown in Figure 5.2.
• Phase 6: Implementation of these local drops in normal stress within the FEM model.
• Phase 7: If the series of the cycles is the last step, go to the next step, else return to
Phase 3.
• Phase 8: Calculation of the new limit load of the pile under monotonic loading after
these cyclic sequences.
247
General conclusions and
perspectives
• General conclusions
The work developed in this thesis starts from the small amount of existing knowledge
about soil-structure interfaces under cyclic loading. The reader should be convicted that the
interface behaviour not only contains the stress limit (Mohr Coulomb or more sophisticated),
but also a kinematic law.
For this reason, a direct shear testing machine for cyclic tests, measuring in 2-D changes in
stress vector and related changes in relative displacement vector (4 variables) is developed.
The direct shear test should be schematised as an element test including two parts: the
interface itself of very small thickness (10 -12 D50 for sands) surrounded by a kind of buffer
zone situated between the interface and the piston of the shear box. In case of direct shear
at constant normal stress, the buffer does not evolve after applying the initial normal stress,
but for constant normal stiffness, the test should be corrected according to the oedometric
compressibility of the buffer. The direct shear test needs another correction due to the loss
of sand between the box and the rough plate, inducing a parasite unexpected contraction.
This shortcoming has been analysed and quantified.
Finally several series of monotonic and cyclic direct shear test between Fontainebleau sand
and two kinds of plates (rough and smooth) were carried out. Only the tests on rough plate
were deeply analysed and modeled. The key parameters are the density index (loose, dense),
the initial normal stress (60 to 310 kPa), the initial stress ratio (shear to normal stress), the
reduced amplitude (change in shear stress to normal stress), the external normal stiffness
(zero for constant normal stress, CNL), and the number of cycles (104 when possible). The
CNL tests were used for identification of parameters, and the CNS tests (constant normal
stiffness > 0) were used for validation as well as special CNL test at variable amplitude.
For each of these tests, a grain size analysis has been performed before and after test, by
a classical sieve method and sometimes using a laser technique. These data have not been
exploited due to the very small difference between original and sheared samples.
A rate type (phenomenological) model of interface behaviour under cyclic loading has been
proposed. Due to the limited class of shear path (non negative prescribed normal stiffness),
an analytical formulation of CNL tests, function of number of cycles, which should be
considered as a pre-integrated interface law has also been proposed. Instead of considering
248
each detail of each cycle, the mean cyclic paths appearing like a pseudo-creep in the sense
of Perzyna, the kernel being empty (pseudo-creep for any value of the parameters) are
described. The single memory parameter is the mean cyclic normal relative displacement
or the interface density under stress. The paths at prescribed normal stiffness (CNS) or CNL
at various amplitude are incrementally integrated, but using finite analytical increments,
which leads to spare a lot of computer time. This kind of integration is using the notion of
the equivalent number of cycles. The step size effect has been studied. It is very important to
notice that direct shear paths at prescribed normal stiffness are ever contractive (decreasing
normal stress), and leading to the rupture (critical state) after a small or large number of
cycles (Ncrit ).
After this study at the constitutive level, the structural studies related to soil-structure
interfaces, using the FEM Plaxis, and the simplest constitutive model: Mohr-Coulomb with
5 parameters have been carried out. The first analysis has been devoted to model the shear
box itself. Since the model is 2D in plane strain, and not 3-D, the defaults (inhomogeneity)
of the direct shear test are clarified, perhaps magnified. Nevertheless the main features of
the CNL and CNS shear tests are reproduced. The second part of our FEM simulations is
related to pile tests. It deals with the centrifuge model pile tests performed at the IFSTTAR
in Nantes, France. The pull-out and compression pile tests, under monotonic and cyclic
loading, in order to develop an alternate methodology compared to the existing ones have
been modeled. The existing ones are essentially the t-z method, and the more classical
method based on well known constitutive equations, integrated step to step within each
cycle. The methodology uses the FEM program as a tool for managing the soil, the pile,
and their equilibrium. The contribution is to introduce the interface cyclic pseudo-creep
here-above developed in the framework of CNS curves function of the number of cycles.
The complete approach would require the integrated curves along CNS paths. A simplified
analysis, used in these computations, refers to the critical number of cycles (Ncrit ) previously
developed and to a supposed law of evolution. Another simplification of this computation
is to apply the drop of normal stress in one phase only, since the cyclic parameters evolve
during this drop. This methodology is developed, but many refinements remain to be done.
A last chapter is devoted to the calculation of real piles subjected to axial cyclic loading.
Within this chapter the steps of modeling are listed according to a limited number of
interface tests, after selecting them by a preliminary FEM simulation. The methodology
exactly conforms to the one developed for our pile calculations.
• Perspectives for future works
The future perspectives opened by this thesis work are numerous. From the experimental
point of view, it would be interesting to link the grain size evolution during cycles with
the evolution of the normal relative displacement. Some tests with more various cyclic
amplitudes would complete the small number of existing ones. The cyclic pseudo-creep is
presently analytically formulated within an excel sheet. It would be useful to translate it in
Fortran or C for using it directly in a FEM code. The analytical formulas describing CNL
paths are using a large number of coefficients, which is not an obstacle, but a simplification
of this analytical formulation is probably possible. At the structural level, calculating a
pile requires several steps of dialogue between Plaxis and the rest of the world. It would
249
be important to made this exchange automatic. The methodology developed here for pile
calculations needs to be improved by a refined mesh, by considering a larger number of
horizontal layers beside the pile, and by integrating the changes in normal tress instead of
proceeding in one step. In addition the complete procedure should be validated on several
real piles.
250
References
Airey, D., Al-Douri, R., & Poulos, H. (1992). Esimation of pile friction degradation from
shearbox tests. ASTM Geotechnical Testing Journal , 15(4), 388-392.
Al-Douri, R., & Poulos, H. (1992). Static and cyclic shear tests on carbonate sands. ASTM
Geotechnical Testing Journal , 15(2), 138-158.
API, A. P. I. (1993). Foundation design. RP 2A-LRFD-Section G, 64-77.
Armand, G., Boulon, M., Papadopoulos, C., Basanou, M., & Vardoulakis, I. (1998). Me-
chanical behaviour of dionysos marble smooth joints: I, experiments. Mechanics of jointed
and Faulted Rock , Rossmanith (ed), Balkema, Rotterdam, 159-164.
Arslan, H. (2005). Finite element study of soil structure interface problem. The Electronic
Journal of Geotechnical Engineering, EJGE , 10 , Bundle C.
Aubry, D., Modaressi, A., & Modaressi, H. (1990). A constitutive model for cyclic behaviour
of interfaces with variable dilatancy. Computers and Geotechnics, 9(1/2), 47-58.
Boulon, M. (1988). Contribution à la mécanique des interfaces sols-structures. applica-

tion au frottement lateral des pieux. Diplome d’Habilitation à Diriger des Recherches,
Université Joseph Fourier, Grenoble, France.
Boulon, M. (1989). Basic features of soil structure interface behavior. Computers and
Geotechnics, 7 , 115-131.
Boulon, M., & Foray, P. (1986). Physical and numerical simulations of lateral shaft friction
along offshore piles in sand. The 3rd Int. Conf. On Numerical Methods in offshore piling,
Nantes, 127-147.
Boulon, M., Garnica, P., & Vermeer, P. (1995). Soil-structure interaction: Fem computa-
tions, mechanics of geomaterial interfaces. Selvadurai A.P.S. and Boulon M.J. (editors),
Elseviers Science B.V., 147-171.
Boulon, M., Ghionna, V., & Mortara, G. (2003). A strain-hardening elastoplastic model for
sand-interface under monotonic and cyclic loading. Mathematical and Computer Mod-
elling, 37 , 623-630.
Boulon, M., & Nova, R. (1990). Modelling of soil-structure interface behaviour, a compar-
ison between elastoplastic and rate type laws. Computer and Geotechnics, 9 , 21-46.
251
Boutrif, A. (1993). Mesure de comportement d’interface sol-structure à la boı̂te de cisaille-
ment direct annulaire et modélisation. Thèse de doctorat, Université Joseph Fourier,
Grenoble, Grenoble, France.
Brinkgreve, R., Broere, W., & Waterman, D. (2008). Plaxis 2d - manual.
D’Aguiar, S., Modaressi-Farahmand-Razavi, A., Lopez-Caballero, F., & Santos, J. (2008).

Soil-structure interface modeling: Application to pile axial loading. The 12th Interna-
tional Association for Computer Methods and Advances in Geomechanics (IACMAG),
Goa, India, 957-965.
D’Aguiar, S., Modaressi-Farahmand-Razavi, A., Santos, J., & Lopez-Caballero, F. (2011).

Elastoplastic constitutive modeling of soil-structure interfaces under monotonic and cyclic
loading. Computers and Geotechnics, 38 , 430-447.
Dejong, J., Randolph, M., & White, D. (2003). Interface load transfer degradation during
cyclic loading: a microscale investigation. Soils and foundations, 43(4), 81-93.
Desai, C., Drumm, E., & Zaman, M. (1985). Elasto-plastic analysis of the interface behavior
between granular media and structure. ASCE JGE , 111(6), 793-815.
Desai, C., & Nagaraj, B. (1988). Modeling for cyclic normal and shear behavior of interfaces.
ASCE JGE , 114(7), 1198-1217.
Fakharian, K., & Evgin, E. (1997). Cyclic simple shear behaviour of sand-steel interfaces
under constant normal stiffness condition. ASCE JGGE , 123(12), 1096-1105.
Flavigny, E., & Nova, R. (1987). Viscous properties of geomaterials. Geomaterials: Con-
stitutive Equations and Modelling, Edited by Félix Darve, Elsevier Applied Science..
Garnica-Anguas, P. (1993). Simulation numérique du frottement entre solides par équations

intégrales aux frontières et modèle d’interface non linéaire. application aux pieux. Thèse
de doctorat, Université Joseph Fourier, Grenoble, Grenoble, France.
Gennaro, V. D., & Frank, R. (2002). Elasto-plastic analysis of the interface behavior
between granular media and structure. Computers and Geotechnics, 29(7), 547-572.
Hammad, W. (1991). Modélisation non linéaire et étude expérimentale des bandes de ci-
saillement dans le sable. Thèse de doctorat, Université Joseph Fourier, Grenoble, Greno-
ble, France.
Hans, J., & Boulon, M. (2003). A new device for investigating the hydro-mechanical
properties of rock joints. Int. J. Numer. Anal. Meth. Geomech, 27 , 513-548.
Heeres, O., Suiker, A., & de Borst, R. (2002). A comparison between the perzyna vis-
coplastic model and the consistency visoplastic model. European Journal of Mechanics
A/Solids, 21 , 1-12.
Hicher, P., & Shao, J. (2002). Modèles de comportement des sols et des roches. Tome 2:
lois incrémentales, viscoplasticité, endommagement, ed Hermès-Lavoisier .
252
Hoteit, N. (1990). Contribution à l’étude du comportement d’interface sable-inclusion et
application au frottement apparent. Thèse de doctorat, Institut National Polytechnique
de Grenoble, France.
Hu, L., & Pu, J. (2004). Testing and modeling of soil-structure interface. ASCE, Journal
of Geotechnical and Geoenvironmental Engineering, 130(8), 851-860.
Johnston, I., Lam, T., & Williams, A. (1987). Constant normal stiffness direct shear testing
for socketed pile design in weak rock. Géotechnique, 37(1), 83-89.
Kaggwa, W., Booker, J., & Carter, J. (1991). Residual strains in calcareous sand due to
irregular cyclic loading. Journal of Geotechnical Engineering, 17(2), 201-218.
Karstunen, M., Yin, Z., Koskinen, M., Leoni, M., & Vermeer, P. (2008). Recent de-
velopments in constitutive modelling of soft clays. The 12th International Association
for Computer Methods and Advances in Geomechanics (IACMAG), Goa, India, Invited
keynote paper .
Kelly, R. (2001). Development of a large diameter ring shear apparatus and its use for
interface testing. Ph.D. Thesis, University of Sydney, Australia.
Kishida, H., & Uesugi, M. (1987). Tests of the interface between sand and steel in the
simple shear apparatus. Géotechnique, 37(1), 45-52.
Liingaard, M., Augustesen, A., & Lade, P. (2004). Characterization of models for time-
dependent behaviour of soils. ASCE International Journal of Geomechanics, 4(3), 157-
177.
Liu, H., & Ling, H. (2008). Constitutive description of interface behaviour including cyclic
loading and particle breakage within the framework of critical state soil mechanics. Int.
J. Numer. Anal. Meth. Geomech, 32 , 1495-1514.
Liu, H., Song, E., & Ling, H. (2006). Constitutive modeling of soil-structure interface
through the concept of critical state soil mechanics. Mechanics Research Communications,
33 , 515-531.
Luong, M. (1980). Stress-strain aspects of cohesionless soils under cyclic and transient load-
ing. International Symposium on Soils under Cyclic and Transient Loading, Zienkiewicz
O.C. (eds); 7-11 January, 315-324.
Messast, S., Boulon, M., Flavigny, E., & Labanieh, S. (2006). Modélisation constitutive
du comportement cyclique des sables en condition drainée. 18ème Congrès Français de
Mécanique, Grenoble, 1-6.
Mortara, G. (2001). An elastoplastic model for sand-structure interface behaviour under

monotonic and cyclic loading. Ph.D. Thesis. Politecnico di Torino, Italy.
Mortara, G., Boulon, M., & Ghionna, V. (2002). A 2-d constitutive model for cyclic interface
behaviour. International Journal for Numerical and Analytical Methods in Geomechanics,
26 , 1071-1096.
253
Mortara, G., Mangiola, A., & Ghionna, V. (2007). Cyclic shear stress degradation and
post-cyclic behaviour from sand-steel interface direct shear tests. Canadian Geotechnical
Journal , 44 , 739-752.
Moutraji, J. (1992). Etude expérimentale et numérique du cisaillement direct silt-structure
application à l’amélioration du frottement latéral dans les sols fins. Thèse de doctorat,
Université Joseph Fourier, Grenoble, France.
Niemunis, A., Whichtman, T., & Triantafyllidis, T. (2005). A high-cycle accumulation
model for sand. Computers and Geotechnics, 32 , 245-263.
Nova, R., & Wood, D. (1979). A constitutive model for sand in triaxial compression.
International Journal for Numerical and Analytical Methods in Geomechanics, 3(3), 255-
278.
Oumarou, T., & Evgin, E. (2005). Cyclic behaviour of a sand-steel plate interface. Canadian
Geotechnical Journal , 42 , 1695-1704.
Owen, D., & Hinton, E. (1980). Finite elements plasticity: Theory and practice. Pineridge
Press Limited .
Papadopoulos, C., Basanou, M., Vardoulakis, I., Boulon, M., & Armand, G. (1998). Me-
chanical behaviour of dionysos marble smooth joints: Ii, constitutive modeling. Mechanics
of jointed and Faulted Rock , Rossmanith (ed), Balkema, Rotterdam, 165-170.
Pastor, M., Zienkiewicz, O., & Leung, K. (1985). Simple model for transient soil loading
in earthquake analysis ii. non-associative model for sands. Int. J. Numer. Anal. Meth.
Geomech., 9(5), 477-498.
Perzyna, P. (1966). Fundamental problems in viscoplasticity. Adv. Appl. Mech., 9 , 243-377.
Puech, A., Foray, P., Boulon, M., & Desrues, J. (1979). Calcul des pieux à l’arrachement à
partir d’un modèle numérique en contraintes effectives-premiers résultats. VII European
Conf. on soils mechanics and foundation engineering, Brighton, 227-232.
Samtani, N., Desai, C., & Vulliet, L. (1996). An interface model to describe viscoplastic
behavior. International Journal for Numerical and Analytical Methods in Geomechanics,
20 , 231-252.
Schlosser, F., & Guilloux, A. (1981). Le frottement dans le renforcement des sols. Revue
Française de Géotechnique, 16 , 65-79.
Shahrour, I., & Rezaie, F. (1997). An elastoplastic constitutive relation for soil-structure
interface under cyclic loading. Computers and Geotechnics, 21(1), 21-39.
Stijn, F., Karg, C., Degrande, G., & Haegeman, W. (2007). A long term accumulation
model for soils under low amplitude cyclic loading. The 7th Colloque National AFPPS ,
Ecole Centrale Paris.
Tabucanon, J., Airey, D., & Poulos, H. (1995). Pile skin friction in sand from constant
normal stiffness tests. ASTM GTJ , 18(3), 350-364.
254
Uesugi, M., & Kishida, K. (1986). Influential factors of friction between steel and dry sands.
Soil and Foundations, 6(2), 33-46.
Uesugi, M., Kishida, K., & Yasunori, T. (1989). Friction between sand and steel under
repeated loading. Soil and Foundations, 29(3), 127-137.
Wichtmann, T. (2005). Explicit accumulation model for non cohesive soils under cyclic
loading. Ph.D. Thesis, Ruhr Universität Bochum, Germany.
Wichtmann, T., Niemunis, A., & Triantafyllidis, T. (2005). Strain accumulation in sand
due to cyclic loading: drained triaxial tests. Soil Dynamics and Earthquake engineering,
25 , 967-979.
Wichtmann, T., Niemunis, A., & Triantafyllidis, T. (2010). Strain accumulation in sand
due to cyclic loading: On the effect of monotonic and cyclic preloading (miner’s rules).
Soil Dynamics and Earthquake engineering, 30 , 736-745.
Yin, Z. (2006). Modélisation du comportement visqueux de l’argile naturelle. XXIVèmes

Rencontres Universitaires de Génie Civil-Prix Jeunes Chercheurs, 1-8.
Yin, Z., & Karstunen, M. (2008). Influence of anisotropy, destructuration and viscosity
on the behavior of an embankment on soft clay. The 12th International Association for
Computer Methods and Advances in Geomechanics (IACMAG), Goa, India, 4728-4735.
Yoshimi, Y., & Kishida, T. (1981). A ring torsion apparatus for evaluating friction between
soil and surfaces. ASTM Geotechnical Testing Journal , 4(4), 145-152.
255
Appendices
256
Appendix A
Analytical model of monotonic

interface tests
A.1 Detail of functions at before-peak phase
y
y peak
θ
x
x peak
[w]
[ w] peak
Figure A.1: Interface direct shear test at before-post peak
For η([w]), σ([w]) and [u]([w]) diagrams, a function defined by a polynomial function of
degree three is given as:
y = A · x3 + B · x2 + C · x + D (A.1)
where
y = η or σ or [u]
ypeak = ηpeak or σpeak or [u]crit
257
x = [w]
The derivative of a polynomial function is a quadratic function as;
y ′ = 3 · A · x2 + 2 · B · x + C (A.2)
if x = 0, y = 0 and y ′ = θ (the initial slope).

if x = xpeak , y = ypeak (the peak value) and y ′ = 0.
D=0
C =θ
y ′ = 3 · A · x2peak + 2 · B · xpeak + θ = 0
ypeak = A · x3peak + B · x2peak + C · xpeak
θ ypeak
A= −2 3 (A.3)
x2peak xpeak
θ ypeak
B = −2 +3 (A.4)
xpeak x2peak
C = θ (the initial slope) (A.5)

0 for η([w]) and [u]([w])
D= (A.6)
σn for σ([w])
A.2 Detail of functions at post-peak phase
For η([w]) diagram, a function is defined by:
y = A · exp(−B · x2 ) (A.7)
where
x = [w] − [w]peak
let us define:

τ τ
y= −
σn σn crit
258
The difference of stress ratio during the peak and critical value is expressed as:

τ τ
yM = − (A.8)
σn peak σn crit
and
A = yM
Considering the point at 2% of yM ,
0.02 · yM = yM · exp(−B · x2 )
then
ln(0.02)
B=− (A.9)
x21
It can be noted that there is not so much difference in the range 1-5% of yM .
y K
K peak
yM
Kcrit 2%(K peak K crit )

Kcrit
0
[ w] peak
>w@
x = [w] ± [w]peak
x1 = 3[w]peak
Figure A.2: Interface direct shear test at post-post peak of η([w])
σ([w]) and [u]([w]) diagrams are defined in the same function as, for instance,:

dσn [w] − [w]peak
= 1 − exp − · β+ (A.10)
d[w] υ
259
1
1− exp
0 [w]
[ w] peak
Figure A.3: Interface direct shear test at post-post peak of σ([w]) and [u]([w])
if
dσn
[w] = [w]peak → =0
d[w]
dσn
[w] = ∞ → = β+
d[w]
The evaluation of υ can be expressed as:

[w] − [w]peak
1 − exp − · β+ = n · β+ (A.11)
υ
where n is the constant which n = 0.95 is held in this study.

Then
[w]peak [w]peak
υ=− = (A.12)
ln(1 − n) 2.9957
From A.10, the evolution of σn is then expressed as:

[w] − [w]peak
σn = β+ [w] − [w] · exp − · (−υ) + C1 (A.13)
υ
by imposing σn = σpeak for [w] = [w]peak , and substituting υ in A.13, the constant C1 is
derived:
[w]peak
C1 = σpeak − β+ [w]peak − β+ · (A.14)
2.9957
260
Therefore, the evolution of σn as a function of [w] is
σn = σpeak + β+ · ([w] − [w]σ peak )

(A.15)

[w]σ peak 2.9957
− β+ · · 1 − exp ([w] − [w]σ peak ) ·
2.9957 [w]σ peak
261
Appendix B
Results for Monotonic CNL and

CNS tests
B.1 Monotonic CNL tests on loose sand (ID0 ≈ 30%)
262
1 100
0.8 80
experimental
model
K WVn
W (kPa)
0.6 60
experimental
model
0.4 40
0.2 20
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
120 0.05
100 0
model model
Vn (kPa)
80 -0.05
[u] (mm)
60 -0.1
40 -0.15
20 -0.2
0 -0.25
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
100
ID0 | 30 % ; CNL ; Vn = 60 kPa
80
experimental
model
W (kPa)
60
40
20
0
0 20 40 60 80 100 120
Vn (kPa)
Figure B.1: Comparison between experimental data and the model prediction, σn = 60
kPa, CNL test on loose sand (ID0 ≈ 30%) with rough plate
263
1 160
0.8
experimental 120
K WVn
model experimental
W (kPa)
0.6 model
80
0.4
40
0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
200 0.05
0
experimental
150
model
Vn (kPa)
-0.05
[u] (mm)
100 -0.1
experimental
model -0.15
50
-0.2
0 -0.25
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
160
ID0 | 30 % ; CNL ; Vn = 120 kPa
120
experimental
W (kPa)
model
80
40
0
0 40 80 120 160 200
Vn (kPa)
264
1 300
250
0.8
experimental
model 200
K WVn
W (kPa)
0.6
150
experimental
0.4 model
100
0.2
50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
500 0.05
0
400
-0.05
Vn (kPa)
experimental
[u] (mm)
300
experimental model
-0.1
model
200
-0.15
100
-0.2
0 -0.25
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
300
ID0 | 30 % ; CNL ; Vn = 310 kPa
W (kPa)
200
experimental
model
100
0
0 100 200 300 400 500
Vn (kPa)
265
B.2 Monotonic CNL tests on dense sand (ID0 ≈ 90%)
1 80
0.8
60
K WVn
experimental
W (kPa)
0.6 model
40
0.4
experimental
model 20
0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
100 0.4
80 0.3
experimental
Vn (kPa)
[u] (mm)
model
60 0.2
40 0.1
experimental
model
20 0
0 -0.1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
80
ID0 | 90 % ; CNL ; Vn = 60 kPa
60
W (kPa)
40
experimental
model
20
0
0 20 40 60 80 100
Vn (kPa)
kPa, CNL test on dense sand (ID0 ≈ 90%) with rough plate
266
1 160
0.8
120
experimental
K WVn
W (kPa)
0.6 model
80
0.4
experimental
model 40
0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
200 0.4
0.3
150 experimental
[u] (mm)
model
0.2
Vn (kPa)
100
experimental 0.1
model
50
0
0 -0.1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
150
ID0 | 90 % ; CNL ; Vn = 120 kPa
100
W (kPa)
experimental
model
50
0
0 50 100 150 200
Vn (kPa)
267
1 300
250
0.8
experimental
K WVn
model 200
W (kPa)
0.6
150
experimental
0.4 model
100
0.2
50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
500 0.25
400
0.15
Vn (kPa)
[u] (mm)
300
experimental
0.05
model
200 experimental
model
-0.05
100
0 -0.15
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
400
ID0 | 90 % ; CNL ; Vn = 310 kPa
300
experimental
W (kPa)
model
200
100
0
0 100 200 300 400 500
Vn (kPa)
268
B.3 Monotonic CNS tests on loose sand (ID0 ≈ 30%)
1 50
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
Vn0 = 60 kPa ; k = 1000 kPa/mm Vn0 = 60 kPa ; k = 1000 kPa/mm
0.8 40
K WVn (-)
model model
0.6 30
W (kPa)
0.4 20
0.2 10
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
80 0.05
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
0
60
[u] (mm)
experimental
Vn (kPa)
model -0.05
40
-0.1
experimental
model
20
-0.15
0 -0.2
0 1 2 3 4 5 6 0 0.4 0.8 1.2 1.6
[w] (mm) [w] (mm)
60
ID0 | 30 % ; CNS ;
Vn0 = 60 kPa ; k = 1000 kPa/mm
50
experimental
model
40
W (kPa)
30
20
10
0
0 10 20 30 40 50 60 70
Vn (kPa)
Figure B.7: Comparison between experimental data and the model prediction, σn0 = 60
kPa, CNS (k = 1000 kPa/mm) test on loose sand (ID0 ≈ 30%) with rough plate
269
1 50
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
0.8 40
K WVn (-)
W (kPa)
model model
0.6 30
0.4 20
0.2 10
0 0
0 1 2 3 4 0 1 2 3 4
[w] (mm) [w] (mm)
100 0.04
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
80 0
experimental
[u] (mm)
Vn (kPa)
model
60 -0.04
experimental
model
40 -0.08
20 -0.12
0 -0.16
0 1 2 3 4 0 0.4 0.8 1.2 1.6
[w] (mm) [w] (mm)
60
ID0 | 30 % ; CNS ;
Vn0 = 60 kPa ; k = 5000 kPa/mm
50
40 experimental
W (kPa)
model
30
20
10
0
0 10 20 30 40 50 60 70
Vn (kPa)
270
1 80
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
0.8
experimental 60
K WVn (-)
model experimental
W (kPa)
0.6 model
40
0.4
20
0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
160 0.05
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
120 0
Vn (kPa)
[u] (mm)
model model
80 -0.05
40 -0.1
0 -0.15
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
120
ID0 | 30 % ; CNS ;
Vn0 = 100 kPa ; k = 1000 kPa/mm
80
W (kPa)
experimental
model
40
0
0 40 80 120 160
Vn (kPa)
271
1 80
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
0.8
60
K WVn (-)
W (kPa)
experimental
0.6 model
40
0.4
experimental
model 20
0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
160 0.05
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
0
120
experimental
Vn (kPa)
[u] (mm)
model -0.05
80
-0.1
experimental
model
40
-0.15
0 -0.2
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
120
ID0 | 30 % ; CNS ;
Vn0 = 100 kPa ; k = 2000 kPa/mm
80
W (kPa)
experimental
model
40
0
0 40 80 120 160
Vn (kPa)
272
1 60
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
50
0.8
experimental
K WVn (-)
model 40
W (kPa)
0.6 experimental
model
30
0.4
20
0.2
10
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
160 0.04
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
120 0
[u] (mm)
experimental
Vn (kPa)
model
80 -0.04
experimental
model
40 -0.08
0 -0.12
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
120
ID0 | 30 % ; CNS ;
Vn0 = 100 kPa ; k = 5000 kPa/mm
80
W (kPa)
experimental
model
40
0
0 40 80 120 160
Vn (kPa)
273
1 300
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
250
0.8
experimental
200
K WVn (-)
model
W (kPa)
0.6
150
experimental
0.4 model
100
0.2
50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
400 0.04
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
300 0
[u] (mm)
experimental
Vn (kPa)
model
200 -0.04
experimental
model
100 -0.08
0 -0.12
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
300
ID0 | 30 % ; CNS ;
Vn0 = 310 kPa ; k = 1000 kPa/mm
200
W (kPa)
experimental
model
100
0
0 100 200 300 400
Vn (kPa)
274
1 250
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
0.8 200
experimental
K WVn (-)
model
W (kPa)
0.6 150
experimental
0.4 model 100
0.2 50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
400 0.04
ID0 | 30 % ; CNS ; ID0 | 30 % ; CNS ;
0
300
experimental
Vn (kPa)
[u] (mm)
model -0.04
200
-0.08
experimental
100 model
-0.12
0 -0.16
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
300
ID0 | 30 % ; CNS ;
Vn0 = 310 kPa ; k = 5000 kPa/mm
200
W (kPa)
experimental
model
100
0
0 100 200 300 400
Vn (kPa)
275
B.4 Monotonic CNS tests on dense sand (ID0 ≈ 90%)
1 160
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
0.8
120
K WVn
W(kPa)
0.6
80
0.4 experimental
experimental
model
model
40
0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
250 0.25
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
200
0.15
Vn (kPa)
150
[u] (mm)
0.05
100 experimental
experimental model
model -0.05
50
0 -0.15
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
200
ID0 | 90 % ; CNS ;
Vn0 = 60 kPa ; k = 1000 kPa/mm
150
W(kPa)
100
50
experimental
model
0
0 50 100 150 200 250
Vn (lkPa)
kPa, CNS (k = 1000 kPa/mm) test on dense sand (ID0 ≈ 90%) with rough plate
276
1 300
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
250
0.8
K WVn
200
W(kPa)
0.6
150
0.4
experimental 100
model experimental
0.2 model
50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
500 0.2
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
400 0.15
experimental
model
Vn (kPa)
[u] (mm)
300 0.1
200 0.05
experimental
model
100 0
0 -0.05
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
400
ID0 | 90 % ; CNS ;
Vn0 = 60 kPa ; k = 5000 kPa/mm
300
W(kPa)
200
experimental
model
100
0
0 100 200 300 400 500
Vn (kPa)
277
1 200
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
0.8 160
K WVn (-)
W (kPa)
0.6 120
0.4 80
model model
0.2 40
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
400 0.2
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
0.15
300
experimental
Vn (kPa)
[u] (mm)
model 0.1
200
0.05
experimental
100 model
0
0 -0.05
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
250
ID0 | 90 % ; CNS ;
Vn0 = 100 kPa ; k = 1000 kPa/mm
200
150
W (kPa)
100
experimental
model
50
0
0 50 100 150 200 250 300
Vn (kPa)
278
1 300
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
250
0.8 experimental
experimental
model
200
K WVn (-)
model
W (kPa)
0.6
150
0.4
100
0.2
50
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
400 0.2
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
0.15
300 experimental
Vn (kPa)
model
[u] (mm)
0.1
200
experimental
0.05
model
100
0
0 -0.05
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
300
ID0 | 90 % ; CNS ;
Vn0 = 100 kPa ; k = 2000 kPa/mm
200
W (kPa)
experimental
model
100
0
0 100 200 300 400
Vn (kPa)
279
1 400
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
0.8
experimental 300
model
W (kPa)
K WVn (-)
0.6
200
0.4 experimental
model
100
0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
600 0.2
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
500
0.15
experimental
Vn (kPa)
400 model
[u] (mm)
0.1
300
experimental
model 0.05
200
0
100
0 -0.05
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
500
ID0 | 90 % ; CNS ;
Vn0 = 100 kPa ; k = 5000 kPa/mm
400
300
W (kPa)
200
experimental
model
100
0
0 100 200 300 400 500 600
Vn (kPa)
280
1 400
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
0.8
experimental 300
model experimental
K WVn (-)
0.6 model
W (kPa)
200
0.4
100
0.2
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
600 0.2
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
500
0.15
experimental
400 model
Vn (kPa)
[u] (mm)
0.1
300
experimental
model 0.05
200
0
100
0 -0.05
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
400
ID0 | 90 % ; CNS ;
Vn0 = 310 kPa ; k = 1000 kPa/mm
300
experimental
model
W (kPa)
200
100
0
0 100 200 300 400 500
Vn (kPa)
281
1 500
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
0.8 400
experimental
K WVn (-)
model
W (kPa)
0.6 300
experimental
model
0.4 200
0.2 100
0 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
800 0.16
ID0 | 90 % ; CNS ; ID0 | 90 % ; CNS ;
0.12
600
0.08
Vn (kPa)
[u] (mm)
400 0.04
experimental
model 0
experimental
200
model
-0.04
0 -0.08
0 1 2 3 4 5 6 0 1 2 3 4 5 6
[w] (mm) [w] (mm)
600
ID0 | 90 % ; CNS ;
Vn0 = 310 kPa ; k = 5000 kPa/mm
500
400
W (kPa)
300
experimental
200
model
100
0
0 100 200 300 400 500 600 700
Vn (kPa)
282
Appendix C
Cyclic interface models
C.1 Constitutive interpolations
• Term of Wc , ”contraction” : 5 conditions are described as:

– 1st condition: for ηcm = 0 → Wc = ā with 0 < ā < 1 (contacting for ηcm = 0)
– 2nd condition: for |ηcm | = η∗ → Wc = 1
– 3rd condition: for |ηcm | = ηpeak → Wc = 0
– 4th condition: for For |ηcm | = η∗ → Wc′ = 1
– 5th condition: for |ηcm | = ηpeak → Wc′ = 0
Wc is expressed as a polynomial function of degree 4 in ηcm ,
2
3 4
Wc = a + b |ηcm | + cηcm + d ηcm + eηcm (C.1)
The first condition gives a = ā.

In case of Wc′ = 0, the 4th and 5th conditions require the multiplicative factors (|ηcm | − η∗ )
and (|ηcm | − ηcm ) with the power ≥ 1, in Wc′ . Therefore, Wc′ is given as:
Wc′ = (|ηcm | − ηpeak )(|ηcm | − ηpeak )(e |ηcm | + f ) (C.2)
And by integrating;
3
e 4 η
Wc = ηcm + cm (f + e (ηpeak + η∗ ))
4 3 (C.3)
η 2
+ cm (eηpeak η∗ − f (ηpeak + η∗ )) + ηcm f ηpeak η∗
2
283
With the 2nd and 3rd conditions, the parameters e and f are expressed as:
" #
4 (1 − ā)(3η∗ − ηpeak ) ā(3ηpeak − η∗ )
e= + 2 (C.4)
" #
2 (1 − ā)ηpeak (2η∗ − ηpeak ) āη∗ (2η∗ − ηpeak )
f =− + 2 (C.5)
• Term of Wd , ”dilation” : the following conditions are described as:
– 1st condition: for ηcm = 0 → Wd = 0
– 2nd condition: for |ηcm | = η∗ → Wd = 0
– 3rd condition: for |ηcm | = ηpeak → Wd = 1
– 4th condition: for ηcm = 0 → Wd′ = 0
– 5th condition: for |ηcm | = ηpeak → Wd′ = 0
Wd is expressed as a polynomial function of degree 4 in ηcm ,
2
3 4
Wd = a + b |ηcm | + cηcm + d ηcm + eηcm (C.6)
In case of Wd = 0 and Wd′ = 0 for ηcm = 0, the 1st and 4th conditions require the multi-
2 in W and η 2 in W .
plicative factor ηcm d cm d
The 2nd condition requires the multiplicative factor (|ηcm | − η∗ ) with the power ≥ 1, in Wd .
Therefore, Wd is given as:
2
Wd = eηcm (|ηcm − η∗ |) (|ηcm | − f ) (C.7)
and then
3 2
Wd′ = e(4 ηcm − 3ηcm (η∗ + f ) + 2ηcm η∗ f ) (C.8)
Then the 3rd and 5th conditions can impose
4ηpeak − 3η∗
f = ηpeak (C.9)
3ηpeak − 2η∗
3ηpeak − 2η∗
e=− 3 (C.10)
ηpeak (ηpeak − η∗ )2
284
C.2 Correlation between ηcm and ηcm2
Two conditions are defined as the following:

ηmax
• if ηpeak is very small → ηcm2 ≈ ηcm
ηmax
• if ηpeak → 1 then ηcm2 → ηpeak
ηpeak
By imposing x = ηηpeak
max
, y = ηηcm2
cm
and ∆ = ηcm − 1. The conditions for defining the
correction (see Figure C.1 ) are:
1. y = 1 for x = 0
2. y = 1 + ∆ for x = 1
3. y ′ = 0 for x = 0
4. y ′ = 0 for x = 1
In addition, it turns out that y has to be forced close to unity for the low values of x.
This is achieved by imposing an additional condition:
5. y = 1 + ∆/3 for x = 2/3
K cm 2
K cm
K peak
' 1
K cm
'/3
1
x1
K max
2/3 1 x
K peak
Figure C.1: Amplifying ηcm in a controlled manner, ηcm2 in the direction of ηpeak
Finally, the interpolation consists of a polynomial of degree 4 (for 5 conditions):
33 29 21
y=− ∆x4 + ∆x3 − ∆x2 + 1 (C.11)
4 2 4
or
285
" 4 3 2 #
33 ηmax 29 ηmax 21 ηmax
ηcm2 = ηcm + (ηpeak − ηcm ) − + − (C.12)
4 ηpeak 2 ηpeak 4 ηpeak
C.3 Curve of pmax (∆η)
To express the maximum of the hyperbolic function with oblique asymptote, let p = y and
∆η = x and the function is expressed as
α
y = ax + b + (C.13)
c−x
where c < 0, and α = −bc for x = 0 and y = 0 (see Figure C.2).
y
b
1
a
ac b 2 abc
2K peak
c
0 bc b
x
c c
a a
Figure C.2: Limit of pmax (∆η) for a given |ηcm |
Then, the above expression can be rewritten as:
bc
y = ax + b − (C.14)
c−x
with the conditions, y = 0 for x = β and y ′ = 0 for x = δ.

Then β corresponds to 2ηpeak and δ corresponds to pmax
b
β =c− (C.15a)
a
r
bc
δ =c+ (C.15b)
a
286
To represent the the hyperbolic function of pmax (∆η), δ is chosen to substitute x in C.14:
r !
bc bc
y =a c+ +b+ q (C.16)
a bc
a
Finally, the hyperbolic function with oblique asymptote is
√
y = ac + b + 2 abc (C.17)
A parabola passing through the origin and having its vertex on pmax (∆η) is proposed as:
p(∆η), |ηcm |) = α(∆η)2 + β∆η + γ (C.18)
when
• passing the origin → γ = 0
• passing the vertex (k) → pmax (∆ηmax )
and then this parabola can be written as
2
pmax = α∆ηmax + β∆ηmax (C.19)
The derivative of C.17 as
dp
= 2α∆η + β (C.20)
d(∆η)
then gives α and β as the following
pmax
α=− 2
(C.21a)
∆ηmax
pmax
β=2 (C.21b)
∆ηmax
Finally, function of p (∆η), |ηcm |) can be expressed as:
pmax pmax
p=− 2
· ∆η 2 + 2 · ∆η (C.22)
∆ηmax ∆ηmax
or it can be rewritten

∆η ∆η
p = pmax · 2− (C.23)
∆ηmax ∆ηmax
287
Appendix D
Cyclic CNL tests
D.1 Cyclic CNL tests on loose sand (ID0 ≈ 30%) with rough
plate
288
0.0 0.0 0
10LL60 10LM60 10LH60
model model -0.05
model
-0.2 -0.2
-0.1
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.4 -0.4 -0.15
-0.2
-0.6 -0.6
-0.25
-0.8 -0.8 -0.3

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-) N (-)
0.2 2.5 16
10LL60
model 2.0
0.1 12
[w]cm (mm)
10LH60
[w]cm (mm)
[w]cm (mm)
1.5
model
0.0 8
10LM60
1.0
model
-0.1 4
0.5
-0.2 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-) N (-)
Figure D.1: Comparison between experimental data and the model prediction, σn = 60
kPa, ∆τ = 10 kPa
0.0 0.0 0.00

20LL60 20LM60 20LH60
model -0.2 model -0.05 model
-0.2
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.4 -0.10
-0.4
-0.6 -0.15
-0.6
-0.8 -0.20
-0.8 -1.0 -0.25

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-) N (-)
0.4 4 8
20LL60 20LM60
model model
0.2 3 6
[w]cm (mm)
20LH60
[w]cm (mm)
[w]cm (mm)
model
0.0 2 4
-0.2 1 2
-0.4 0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-) N (-)
kPa, ∆τ = 20 kPa
289
0.00 0 0.00
10LL310 10LM310 10LH310
-0.05
-0.04 model -0.05 model model
[u]cm (mm)
[u]cm (mm)
-0.10
[u]cm (mm)
-0.08 -0.1
-0.15
-0.12 -0.15
-0.20
-0.16 -0.2 -0.25
-0.20 -0.25 -0.30

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 1000 2000 3000 4000 5000
N (-) N (-) N (-)
0.2 0.4 8
10LL310 10LH310
model model
0.1 0.3 6
[w]cm (mm)
[w]cm (mm)
[w]cm (mm)
0.0 0.2 4
10LM310
model
-0.1 0.1 2
-0.2 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 1000 2000 3000 4000 5000
N (-) N (-) N (-)
kPa, ∆τ = 10 kPa
0 0 0
20LM310
20LL310 20LH310
-0.04 model -0.05
model -0.1 model
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.08 -0.1
-0.2
-0.12 -0.15
-0.3
-0.16 -0.2
-0.2 -0.4 -0.25

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600 2000
N (-) N (-) N (-)
0.2 0.8 4
20LL310 20LM310 20LH310
model model model
0.1 0.6 3
[w]cm (mm)
[w]cm (mm)
[w]cm (mm)
0.0 0.4 2
-0.1 0.2 1
-0.2 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600 2000
N (-) N (-) N (-)
kPa, ∆τ = 20 kPa
290
D.2 Cyclic CNL tests on dense sand (ID0 ≈ 90%) with rough
plate
0.00 0.0 0.20

10DL60 10DM60
10DH60
model model 0.15 model
-0.05 -0.1
[u]cm (mm)
[u]cm (mm)
0.10
-0.10 [u]cm (mm) -0.2
0.05
-0.15 -0.3
0.00
-0.20 -0.4 -0.05

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 500 1000 1500 2000
N (-) N (-) N (-)
0.2 0.8 16
10DL60
model
0.1 0.6 12
10DH60
[w]cm (mm)
[w]cm (mm)
[w]cm (mm)
model
0.0 0.4 8
10DM60
model
-0.1 0.2 4
-0.2 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 500 1000 1500 2000
N (-) N (-) N (-)
kPa, ∆τ = 10 kPa
0.00 0.0 0.20

20DM60
20DL60 20DH60
-0.05 model 0.15
model -0.1 model
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.10 0.10
-0.2
-0.15 0.05
-0.3
-0.20 0.00
-0.25 -0.4 -0.05

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 200 400 600 800 1000
N (-) N (-) N (-)
0.2 1.6 4
20DL60
model
0.1 1.2 3
20DM60 20DH60
[w]cm (mm)
[w]cm (mm)
[w]cm (mm)
model model
0.0 0.8 2
-0.1 0.4 1
-0.2 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 200 400 600 800 1000
N (-) N (-) N (-)
kPa, ∆τ = 20 kPa
291
0 0.00 0.16
10DL120 10DM120 10DH120
model -0.04 model 0.12 model
-0.04
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.08 0.08
-0.08
-0.12 0.04
-0.12
-0.16 0.00
-0.16 -0.20 -0.04

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 200 400 600 800
N (-) N (-) N (-)
0.2 0.8 4
10DL120 10DM120
model model
0.1 0.6 3
10DH120
[w]cm (mm)
[w]cm (mm)
[w]cm (mm)
model
0.0 0.4 2
-0.1 0.2 1
-0.2 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 200 400 600 800
N (-) N (-) N (-)
kPa, ∆τ = 10 kPa
0.00 0 0.12
20DL310 20DM310 20DH310
model model model
-0.02 -0.04 0.08
[u]cm (mm)
[u]cm (mm)
[u]cm (mm)
-0.04 -0.08 0.04
-0.06 -0.12 0.00
-0.08 -0.16 -0.04

0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-) N (-)
0.2 0.4 3
20DL310 20DM310
20DH310
model model
0.1 0.3 model
[w]cm (mm)
2
[w]cm (mm)
[w]cm (mm)
0.0 0.2
1
-0.1 0.1
-0.2 0.0 0
0 2000 4000 6000 8000 10000 12000 0 2000 4000 6000 8000 10000 12000 0 400 800 1200 1600
N (-) N (-) N (-)
kPa, ∆τ = 20 kPa
292
Appendix E
Cyclic CNS tests
E.1 Cyclic CNS tests on loose sand (ID0 ≈ 30%) with rough
plate
293
160 0.8
20LM100_1000k 20LM100_1000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 1 2 3 4 0 1 2 3 4
N (-) N (-)
0.04 160
20LM100_1000k 20LM100_1000k
model model
120
[u]cm (mm)
Vn cm (kPa)
0
80
-0.04
40
-0.08 0
0 1 2 3 4 -0.08 -0.04 0 0.04
N (-) [u]cm (mm)
Figure E.1: Comparison between experimental data and the model prediction, k = 1000
kPa/mm, σn0 = 100 kPa, ηcm0 = 0.30, ∆τ = 20 kPa
160 0.8
10LM100_2000k 10LM100_2000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 2 4 6 8 0 2 4 6 8
N (-) N (-)
0.04 160
10LM100_2000k 10LM100_2000k
model model
0 120
[u]cm (mm)
Vn cm (kPa)
-0.04 80
-0.08 40
-0.12 0
0 2 4 6 8 -0.12 -0.08 -0.04 0 0.04
N (-) [u]cm (mm)
294
160 0.8
10LL100_5000k 10LL100_5000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 2 4 6 8 10 0 2 4 6 8 10
N (-) N (-)
0.02 160
10LL100_5000k 10LL100_5000k
model model
0 120
[u]cm (mm)
Vn cm (kPa)
-0.02 80
-0.04 40
-0.06 0
0 2 4 6 8 10 -0.06 -0.04 -0.02 0 0.02
N (-) [u]cm (mm)
160 0.8
20LL100_5000k 20LL100_5000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 1 2 3 4 0 1 2 3 4
N (-) N (-)
0.02 160
20LL100_5000k 20LL100_5000k
model model
0 120
[u]cm (mm)
Vn cm (kPa)
-0.02 80
-0.04 40
-0.06 0
0 1 2 3 4 -0.06 -0.04 -0.02 0 0.02
N (-) [u]cm (mm)
295
E.2 Cyclic CNS tests on dense sand (ID0 ≈ 90%) with rough
plate
100 1
10DM60_1000k 10DM60_1000k
80 model 0.8 model
Vn cm (kPa)
60 0.6
K cm (-)
40 0.4
20 0.2
0 0
0 400 800 1200 1600 0 400 800 1200 1600
N (-) N (-)
0.02 100
10DM60_1000k 10DM60_1000k
model 80 model
0
[u]cm (mm)
Vn cm (kPa)
60
-0.02
40
-0.04
20
-0.06 0
0 400 800 1200 1600 -0.06 -0.04 -0.02 0 0.02
N (-) [u]cm (mm)
100 1
10DH60_1000k 10DH60_1000k
80 model 0.8 model
Vn cm (kPa)
K cm (-)
60 0.6
40 0.4
20 0.2
0 0
0 50 100 150 200 0 50 100 150 200
N (-) N (-)
0.02 100
10DH60_1000k 10DH60_1000k
model 80 model
0
Vn cm (kPa)
[u]cm (mm)
60
-0.02
40
-0.04
20
-0.06 0
0 50 100 150 200 -0.06 -0.04 -0.02 0 0.02
N (-) [u]cm (mm)
296
160 1
10DH100_1000k 10DH100_1000k
model model
0.8
120
Vn cm (kPa)
0.6
K cm (-)
80
0.4
40
0.2
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
N (-) N (-)
0.01 160
10DH100_1000k 10DH100_1000k
model model
0 120
Vn cm (kPa)
[u]cm (mm)
-0.01 80
-0.02 40
-0.03 0
0 1000 2000 3000 4000 5000 -0.03 -0.02 -0.01 0 0.01
N (-) [u]cm (mm)
160 0.8
20DM100_1000k 20DM100_1000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
N (-) N (-)
0.02 160
20DM100_1000k 20DM100_1000k
model model
0 120
[u]cm (mm)
Vn cm (kPa)
-0.02 80
-0.04 40
-0.06 0
0 1000 2000 3000 4000 5000 -0.06 -0.04 -0.02 0 0.02
N (-) [u]cm (mm)
297
160 0.8
10DM100_2000k 10DM100_2000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 2000N (-) 4000 6000 8000 0 2000 4000 6000 8000
N (-)
0.02 160
10DM100_2000k 10DM100_2000k
model model
120
0
[u]cm (mm)
Vn cm (kPa)
80
-0.02
40
-0.04 0
0 2000 4000 6000 8000 -0.04 -0.02 0 0.02
N (-) [u]cm (mm)
160 0.8
10DH100_2000k 10DH100_2000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 1000 2000 3000 4000 0 1000 2000 3000 4000
N (-) N (-)
0.02 160
10DH100_2000k 10DH100_2000k
model model
0.01 120
Vn cm (kPa)
[u]cm (mm)
0 80
-0.01 40
-0.02 0
0 1000 2000 3000 4000 -0.02 -0.01 0 0.01 0.02
N (-) N (-)
kPa/mm, σn0 = 100 kPa, ηcm0 = 0.50, ∆τ = 10 kPa, rough plate on dense sand (ID0 ≈ 90%)
298
160 0.8
10DM100_5000k 10DM100_5000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
N (-) N (-)
0.01 160
10DM100_5000k
model
0 120
[u]cm (mm)
Vn cm (kPa)
-0.01 80
10DM100_5000k
model
-0.02 40
-0.03 0
0 500 1000 1500 2000 -0.03 -0.02 -0.01 0 0.01
N (-) [u]cm (mm)
160 0.8
10DH100_5000k 10DH100_5000k
model model
120 0.6
Vn cm (kPa)
K cm (-)
80 0.4
40 0.2
0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
N (-) N (-)
0.01 160
10DH100_5000k
model
0 120
[u]cm (mm)
Vn cm (kPa)
-0.01 80
10DH100_5000k
model
-0.02 40
-0.03 0
0 500 1000 1500 2000 -0.03 -0.02 -0.01 0 0.01
N (-) [u]cm (mm)
299
400 0.8
10DM310_1000k 10DM310_1000k
model model
300 0.6
Vn cm (kPa)
K cm (-)
200 0.4
100 0.2
0 0
0 10000 20000 30000 40000 0 10000 20000 30000 40000
N (-) N (-)
0.04 400
10DM310_1000k
model
0 300
[u]cm (mm)
Vn cm (kPa)
-0.04 200
10DM310_1000k
model
-0.08 100
-0.12 0
0 10000 20000 30000 40000 -0.12 -0.08 -0.04 0 0.04
N (-) [u]cm (mm)
400 0.8
300 0.6
Vn cm (kPa)
K cm (-)
200 0.4
10DH310_1000k 10DH310_1000k
model model
100 0.2
0 0.0
0 5000 10000 15000 20000 0 5000 10000 15000 20000
N (-) N (-)
0.02 400
10DH310_1000k
model
0.00 300
[u]cm (mm)
Vn cm (kPa)
-0.02 200
10DH310_1000k
model
-0.04 100
-0.06 0
0 5000 10000 15000 20000 -0.06 -0.04 -0.02 0 0.02
N (-) [u]cm (mm)
300
Appendix F
Résumé français étendu de la

Thèse
301
302
Titre de la thèse
Essais et modélisation du cisaillement cyclique sol-structure à grand nombre de cycles.

Application aux pieux
Sommaire
Introduction générale
Chapitre 1. Etude bibliographique des interfaces sol-structure sous chargement cyclique
Chapitre 2. Essais de cisaillement direct d’interfaces
2.1. La machine d’essai, les mesures
2.2. Les matériaux, sable et plaque rugueuse
2.3. Mode opératoire
2.4. Correction du biais expérimental
2.5. Interprétation des essais
2.5.1. Poids volumique et épaisseur d’interface
2.5.2. Pseudo-fluage cyclique et chemin cyclique moyen
2.6. Programme d’essais et quelques résultats typiques
2.6.1. Programme d’essais
2.6.2. Résultats typiques
Chapitre 3. Modélisation constitutive des essais de cisaillement direct
3.1. Un modèle phénoménologique « pré-intégré »
3.2. Modélisation des essais monotones
3.3. Modélisation des essais cycliques CNL d’identification
3.3.1. Formulation analytique des chemins clés CNL de cisaillement cyclique
3.3.2. Interpolation constitutive
3.3.3. Résultats comparés à l’expérience
3.4. Modélisation d’essais CNL à amplitude cyclique variée
3.4.1. Nombre équivalent de cycles. Validation sur essais CNL à amplitudes variées
3.4.2. Modélisation de chemins CNL possibles de type tempête
3.5. Modélisation des essais cycliques CNS de validation
3.5.1. Intégration par « incréments finis », pré-intégrés
Chapitre 4. Modélisation par la méthode des éléments finis
4.1. Introduction
4.2. Modélisation 2-D de l’essai de cisaillement direct sol-structure monotone à la boîte de cisaillement
4.3. Modélisation des sollicitations cycliques à grand nombre de cycles
4.3.1. Pieu modèle centrifugé en traction, monotone et cyclique
4.3.2. Pieu modèle centrifugé en compression, monotone et cyclique
Chapitre 5. Calcul de pieux réels chargés cycliquement, pour les projets de génie civil
Conclusions générales et perspectives
Annexes
Introduction générale
La problématique des interfaces sol-structure est communément présente dans de nombreux

problèmes géotechniques, fondations superficielles, fondations sur pieux, parois moulées,
renforcement des sols, oléoducs, tunnels… Le récent développement des installations liées à
l’énergie, renouvelable ou non (plateformes off-shore, éoliennes, hydroliennes,…) amène les
professionnels à s’intéresser de manière accrue aux effets des sollicitations cycliques à grand
nombre de cycles, d’autant qu’aucune recommandation nationale ni internationale ne permet
de quantifier la dégradation cyclique, les laissant sans guide dans leurs travaux de conception.
303
La thèse, tentant d’apporter une contribution à ce vide scientifique et technique, est composée
de 5 chapitres, comme indiqué au sommaire ci-dessus.
Le premier chapitre fait une revue bibliographique des travaux expérimentaux et de
modélisation de la littérature, caractérisant et quantifiant la notion d’accumulation cyclique
lors du cisaillement direct cyclique à contrainte normale constante et à rigidité normale
imposée.
Le second chapitre décrit tout d’abord la machine de cisaillement direct développée dans le
cadre de la thèse, permettant de réaliser et d’enregistrer automatiquement un grand nombre de
cycles (typiquement 10 000), avec l’asservissement adéquat permettant de suivre n’importe
quel chemin de cisaillement. L’ensemble et le détail des cycles des essais de cisaillement,
entre sable de Fontainebleau et plaque rugueuse ou lisse, à contrainte normale constante ou à
rigidité normale imposée, est enregistré, puis interprété.
Au troisième chapitre, relatif à la modélisation constitutive, est exposée une formulation
analytique des essais de cisaillement direct monotones (à contrainte normale constante et à
rigidité normale imposé) et cycliques à contrainte normale constante, ainsi que des
compressibilités. Il est largement fait appel à l’interpolation, sur la densité, les contraintes
normales, les niveaux cycliques moyens, et l’amplitude cyclique. Dans un second temps, les
essais cycliques à rigidité normale imposée sont modélisés par incréments analytiques finis
basés sur les formulations précédentes, la variable de mémoire étant la densité d’interface. De
même pour des essais à amplitude variée.
Le quatrième chapitre est consacré à la modélisation par éléments finis, utilisant le logiciel
PLAXIS. C’est tout d’abord une modélisation de l’essai de cisaillement direct sable-plaque
rugueuse monotone lui-même, en 2-D déformation plane, dans une boite légèrement
rugueuse, avec un sable lâche ou dense, et selon un chemin à contrainte normale constante ou
à rigidité normale imposée. Les hétérogénéités sont bien identifiées, mais globalement, les
chemins sont conformes à l’expérience. Puis sont modélisés 2 essais de pieu modèle
centrifugés sollicités cycliquement, l’un en traction, l’autre en compression. La dégradation
cyclique à l’interface sable-pieu est simulée à partir de la modélisation à rigidité normale
imposée développée au troisième chapitre.
Le cinquième et dernier chapitre propose une méthodologie, basée sur les résultats des
troisième et quatrième chapitre, permettant de modéliser, moyennant un petit nombre d’essais
et un temps de calcul acceptable, le comportement de pieux réels chargés cycliquement.
Des conclusions générales sont tirées de ces travaux, et les perspectives de travaux ultérieurs
sont proposées.
Chapitre 1. Etude bibliographique des interfaces sol-structure sous chargement cyclique
Dans cette revue bibliographique, on décrit d’abord le concept de cisaillement direct, essai
élémentaire (homogène ? comment ?) simulant le comportement mécanique d’une fraction du
contact sol-structure, et les dispositifs expérimentaux développés par les chercheurs à cette
fin : cisaillement plan unidirectionnel, cisaillement(s) annulaire(s), cisaillement plan bi-
directionnel. La question de l’épaisseur d’interface est abordée et discutée. Puis vient une
présentation des variables mesurables d’interface, à comparer aux variables utilisées pour
rendre compte du comportement volumique des sols : les déplacements relatifs qui tenant lieu
des déformations du comportement volumique, et les composantes de vecteur contrainte,
analogues des composantes de tenseur contrainte du comportement volumique. Puis sont
évoqués les facteurs classiquement mis en avant, qui influent sur le comportement des
interfaces sol-structure (sols essentiellement granulaires) : la compacité du sol, la rugosité de
la plaque censée simuler la structure, le niveau de contrainte, la résistance des grains (sable
304
siliceux versus sable carbonaté), et enfin le chemin de cisaillement direct. Sont soulignés le
rôle et la représentativité de tout premier ordre des chemins de cisaillement direct à rigidité
normale imposée (externe, ne pas confondre avec la rigidité normale interne, ou
compressibilité, d’une interface). De même, la contractance, toujours observée pour un petit
nombre de cycles sur ces chemins, est rapportée. S’agissant de sollicitations cycliques (le
sujet de cette thèse), les publications abordant le cisaillement direct et les sollicitations
cycliques sont en nombre fort restreint, voire inexistantes lorsqu’on examine le grand nombre
de cycles. Par ailleurs, les essais réalisés concernent généralement de grands cycles. En
l’absence de résultats couvrant bien le champ de cette thèse, nous avons cité les travaux de
grande qualité produits par Wichtmann, concernant une série consistante d’essais triaxiaux
cycliques sur sable, et nous nous en sommes largement inspirés (figure 1). Accumulation de
déformation a pour cet auteur le sens de déformation irréversible après cycles.
La seconde partie de ce chapitre est consacrée aux lois constitutives d’interfaces sol-structure.
Sont évoqués les modèles phénoménologiques et les modèles élasto-plastiques, ainsi que les
modèles de pseudo-fluage cyclique. Le formalisme du pseudo-fluage cyclique est issu des
travaux fondateurs de Perzyna, sur son modèle élasto-visco-plastique. Hormis l’élasticité du
matériau, ce modèle fait appel à trois ingrédients : une loi d’écoulement visco-plastique, un
noyau visqueux, défini à partir de l’espace des contraintes, séparant les zones à et sans
déformations visco-plastiques, et enfin un paramètres scalaire de viscosité.
Figure 1. Essais triaxiaux cycliques : a) accumulation de la norme de déformation εacc

fonction du nombre de cycles, pour différents taux de déviateur cyclique moyen. b) relation
entre déformation déviatoire accumulée (εacc q) et déformation volumique accumulée (εacc v)
selon le taux de déviateur cyclique moyen (εacc v >0 contractant), d’après Whichtmann (2005).
Chapitre 2. Essais de cisaillement direct d’interfaces
2.1. La machine d’essai, les mesures
La machine d’essai est une boîte de cisaillement direct traditionnelle, modifiée pour répondre
aux besoins de l’étude. La masse morte génératrice de la contrainte normale a été remplacée
305
par un moteur électrique double sens, asservi de manière à respecter la consigne
correspondant au chemin de cisaillement direct souhaité (figure 2). La demi-boîte inférieure a
été remplacée par une plaque rugueuse simulant la rugosité de la structure en contact avec le
sol (situé dans la demi-boîte supérieure). Les forces normale et de cisaillement sont mesurée
par 2 capteurs de force équipés de jauges de déformation, dûment étalonnés. Les
déplacements relatifs sol/plaque sont mesurés par 2 LVDT également dûment étalonnés, l’un
pour le déplacement relatif normal, l’autre pour le déplacement relatif tangentiel. En outre, un
LVDT supplémentaire mesure le déplacement vertical de la demi-boîte supérieure, à titre de
contrôle. Une carte analogique digitale permet la communication entre la machine et
l’ordinateur (asservissement et mesure).L’essai est piloté par un programme idoine, écrit en
langage C, ,qui affiche à l’écran les valeurs utiles en temps réel. L’ensemble des mesures est
acquis et stocké dans un fichier texte. Ce dispositif permet de réaliser un cisaillement direct
monotone ou cyclique. Dans les deux cas, la contrainte normale initiale est appliquée à vue.
Ensuite le chemin de cisaillement direct est choisi. Dans le cas cyclique, l’opérateur choisit 2
seuils de contrainte de cisaillement l’un maximum et l’autre une minimum. Enfin le
cisaillement est déclenché et la suite de l’essai est alors automatique, à vitesse relative
tangentielle choisie au préalable et constante, éventuellement inversée (automatiquement) lors
des cycles. On peut enchaîner, sans rupture des mesures, plusieurs séquences, telle que
variation de contrainte normale, et/ou cisaillement.
Figure 2. La boîte de cisaillement direct modifiée, le moteur générateur de la contrainte

normale, les 2 capteurs de force (normale et tangentielle).
Les variables mesurées sont, en fonction du temps t, les déplacements relatifs sol/plaque
normal [u (t )] et tangentiel [w(t )] , les contraintes normale σ n (t ) et de cisaillement τ (t ) , le
déplacement relatif vertical boîte/plaque. Le déplacement relatif tangentiel [w(t )] est toujours
imposé (t = temps). Les chemins possibles sont le chargement et le déchargement pseudo-
oedométrique (en contrainte normale), et le cisaillement résultant de la consigne (1), soit :
306
C = Δσ n − kΔ[u ] (1)
- à contrainte normale constante (k = 0)

- à rigidité normale imposée (k > 0)
- à volume constant (encore appelés à dilatance empêchée) (k → ∞)
L’asservissement est généralement réalisé avec un écart Δσ n = ±5kPa .
2.2. Les matériaux, sable et plaque rugueuse
Le sable testé est le sable de Fontainebleau standard NE34, réputé « uniforme » en raison de
son Cu ,utilisé dans le cadre SOLCYP, dont les propriétés physiques sont indiquées tableau 1,
et dont la courbe granulométrique est représentée figure 3. Les plaques rugueuses simulant le
Tableau 1. Propriétés physiques du sable de Fontainebleau NE34.
Figure 3. Sable de Fontainebleau, aspect et courbe granulométrique.
matériau structural (le pieu par exemple) sont de 2 types. La plaque « rugueuse »est une
plaque rigide d’acier (épaisseur 1 cm) sur laquelle a été collé, à l’araldite, du sable de
Fontainebleau NE34. Cette plaque est restaurée tous les 10 essais, compte tenu de
l’arrachement éventuel de grains, réduisant progressivement sa rugosité. L’étude
morphologique de ces plaques a conduit aux paramètres R max = 0.2 mm, ou R n = 0.87, ce qui
les classe comme « très rugueuses ». Le second type de plaque est la même plaque d’acier,
non couverte par du sable, et dite « plaque lisse » (R max = 14 μm, ou R n = 0.06).
2.3. Mode opératoire
Nous avons testé 2 types d’échantillons sable-structure : avec sable dense et avec sable lâche.
Dans tous les cas, la préparation de l’échantillon est délicate, et faite en insérant un clinquant
307
de laiton de 0.30 mm d’épaisseur entre la plaque et la boîte (figure 4), compte tenu du d50 du
sable (0,23 mm), et du fait que la boîte se rapproche de la plaque lors de l’application de la
contrainte normale nominale initiale relative à l’essai. La densité faible (ID0 ≈ 30 %) est
obtenue par déversement du sable d’une faible hauteur de chute, tandis que la densité élevée
(ID0 ≈ 90 %) nécessite un compactage à l’aide d’une aiguille vibrante. Chaque essai est
constitué des phases successives suivantes, après préparation de l’échantillon:
- Montée en contrainte normale jusqu’à la contrainte normale initiale
- Réglage des paramètres de l’essai
- Cisaillement direct selon le chemin choisi
- Arrêt du cisaillement par l’opérateur au terme du but de l’essai (cas monotone :
déplacement relatif tangentiel maximum ; cas cyclique, nombre maximum de cycles
ou, si « rupture » avant terme, déplacement relatif tangentiel maximum atteint (≈ 15
mm).
- Dans le cas cyclique, cisaillement final de grande amplitude, afin d’évaluer la
résistance de l’interface après les cycles.
- Réduction de la contrainte normale jusqu’à 0 kPa
- Arrêt de l’essai proprement dit
- Démontage du piston de la boîte de cisaillement
- Extraction de la partie supérieure de l’échantillon de sable
- Prélèvement de la zone d’interface ( épaisseur ≈ 10 à 12 d50, échantillon Ef1)
- Prélèvement et pesage du sable « perdu » entre la demi-boîte inférieure et la plaque au
cours de l’essai (échantillon Ef2, en quantité très faible !).
- Analyse granulométrique des échantillons Ef1 et Ef2.
Figure 4. Dispositions d’espacement boîte/plaque lors de la préparation de l’échantillon

d’interface sable-structure
2.4. Correction du biais expérimental
Le d50 du sable étant de l’ordre de 0,23 mm, et l’espacement initial boîte/plaque, hors
contrainte normale, étant de 0,30 mm, on peut imaginer qu’il existe une « perte de sable » au
cours du cisaillement, malgré l’uniformité importante de granulométrie du sable, et le
rapprochement boîte/plaque lors de l’application de la contrainte normale initiale
(entrainement de la demi-boîte inférieure par frottement sable/boîte). Ce biais, faible mais
mesurable, rapporté par de nombreux auteurs, est inhérent à l’essai de cisaillement direct
308
matériau granulaire/structure. Il crée une contractance fictive parasite qui doit absolument être
quantifiée et doit corriger le déplacement relatif normal [u ] . Nous disposons de la masse
finale de sable « perdu », en fin d’essai (§ 2.3). Il s’agit alors d’affecter localement (entre
début et fin de cisaillement) ce biais global. Entre l’affectation proportionnelle au temps
courant de l’essai, et l’affectation proportionnelle au déplacement relatif tangentiel cumulé en
valeur absolue, nous avons choisi la seconde option, semblant plus physique. Les résultats
d’essais sont donc présentés corrigés en ce qui concerne le déplacement relatif normal [u (t )] .
2.5. Interprétation des essais
2.5.1. Poids volumique et épaisseur d’interface
La zone d’interface étant d’épaisseur très faible (on admet généralement ≈ 5 à 10 d50, Boulon
(1988), Hoteit (1990)) par rapport à l’épaisseur d’échantillon de sable (≈ 2 cm), nous avons
résolument interprété l’échantillon de sable comme formé de 2 couches séparément
homogènes : la couche d’interface, au contact de la structure, et le reste de l’échantillon,
couche « tampon » en quelque sorte. La couche tampon n’est pas soumise au cisaillement, elle
ne subit que des déformations oedométriques sous l’effet des seules variations de contrainte
normale. La couche d’interface est par contre soumise au chargement dans son ensemble,
variation de contrainte normale et variation de contrainte de cisaillement (figure 5). Ceci n’est
bien entendu qu’un schéma, l’épaisseur étant une valeur moyenne pour l’échantillon. Se pose
alors la question de la détermination de l’épaisseur de cette zone d’interface : comment
identifier cette épaisseur ? Nous avons utilisé les essais de cisaillement direct monotones
CNL, à plusieurs niveaux de contrainte normale, et à plusieurs densités initiales. L’idée
directrice est que l’épaisseur d’interface est peu dépendante de la densité initiale et du niveau
de contrainte normale, du moins tant que les grains ne sont pas broyés. Par ailleurs la densité
critique doit être atteinte dans l’interface lors de grands déplacements relatifs tangentiels.
Figure 5. Hypothèse des zones séparément homogènes dans l’échantillon de sol (épaisseur h)
situé dans la boîte de cisaillement : au contact de la structure , l’interface (épaisseur t), au
dessus, le « tampon » oedométrique (épaisseur h-t).
Les figures 6 et 7 montrent le test de cette hypothèse, qui se révèle fructueuse. Néanmoins,
l’épaisseur d’interface semble dépendre de la densité initiale (grains plus libres de leurs
mouvements à densité initiale faible ?). Nous retiendrons pour l’interprétation de nos essais,
309
les valeurs indiquées au tableau 2. On retiendra pour la suite que les poids volumiques
indiqués sont des poids volumiques sous contrainte, résultant du poids volumique intial
γ d 0 et du déplacement relatif normal d’interface [u ] .
Figure 6. Poids volumique d’interface (cas plaque rugueuse) selon plusieurs hypothèses
d’épaisseur d’interface : a) densités initiales faible et forte confondues. b) valeurs optimales
différentes pour densités initiales faible et forte.
Figure 7. Déplacement relatif normal et poids volumique d’interface (cas plaque lisse) selon
plusieurs hypothèses d’épaisseur d’interface : a) déplacement relatif (échantillon et interface)
densités initiales faible et forte. b) valeurs optimales pour densités initiales faible et forte.
Plaque ID0 = 90 % ID0 = 30 %

Rugueuse 10 d50 12 d50
lisse 3 d50 9 d50
Tableau 2. Epaisseurs d’interface identifiées par examen des valeurs critiques sur chemins
CNL, selon la densité initiale et la contrainte normale initiale.
310
2.5.2. Pseudo-fluage cyclique et chemin cyclique moyen
L’idée de base ayant prévalu à l’origine de cette étude a été de considérer les déplacements
relatifs irréversibles accumulés au cours des cycles comme un pseudo-fluage cyclique, le
nombre de cycles tenant lieu de temps fictif. C’est la raison pour laquelle nos essais ont été
(pré)interprétés en fonction de la notion de chemin cyclique moyen. Les essais sont ainsi
présentés interprétés en termes de variables déplacement relatif cyclique moyen normal [u ]cm
et tangentiel [w]cm , en fonction du nombre de cycles ( ou = 10000 cycles) et des autres
paramètres (§ 2.6.1). L’évolution des rigidités normale et tangentielle, de la résistance
d’interface, et de la granulométrie est également visée.
2.6. Programme d’essais et quelques résultats typiques
2.6.1 Programme d’essais
Le programme d’essais de base couvre, conformément au § 2.2, 2 types de contacts (rugueux

et lisse), et 2 types de chemins, à contrainte normale constante (CNL, k = 0, en vue de
l’identification des paramètres constitutifs), et à rigidité normale imposée (CNS, k > 0, en vue
de la validation du modèle constitutif), pour chacun 10 000 cycles, sauf cas de rupture
prématurée. La définition des paramètres de ces essais est basée sur la notion de « plans
d’expérience ». Les tableaux suivants (3 à 5) donnent les paramètres clés de ces essais, à
savoir :
- l’indice initial de densité ID0
- la contrainte normale cyclique moyenne initiale σ ncm 0
- les seuils de contrainte de cisaillement ( τ min < τ < τ max )
- le rapport de contraintes cycliques moyennes (cisaillement à normale) η cm 0
Δτ
- l’amplitude cyclique ( Δτ = τ max − τ min , Δη 0 = )
σ ncm0
En outre, un petit nombre d’essais CNL complémentaires, à amplitude cyclique variée au
cours de l’essai, ont été menés à bien, aux fins de validation tableau 6)..
2.6.2. Résultats typiques
Le déplacement relatif tangentiel reste identiquement nul (aux parasites d’acquisition près)
lorsque les cycles sont centrés (ηcm 0 = 0 ). Il est hautement remarquable, pour les applications,
que les essais CNS sont toujours contractants, ce qui explique le concept traditionnel
(souvent faussement interprété) de dégradation cyclique (ce n’est pas le coefficient de
frottement qui est dégradé, croyance courante facile (dégradation de rugosité), mais le niveau
de contrainte normale, ce qui est plus difficile à conceptualiser!).
311
Tableau 3. Série d’essais CNL,
plaque rugueuse
Tableau 4. Série d’essais CNS,

plaque rugueuse
312
Tableau 5. Série d’essais
CNL et CNS, plaque lisse
ID0 (%)
σ ncm0 (kPa) η cm0 Δτ (kPa) N
30 120 0,3 10 puis 20 5000 puis 5000

30 120 0,3 20 puis 10 5000 puis 5000
90 120 0,3 10 puis 20 5000 puis 5000
90 120 0,3 20 puis 10 5000 puis 5000
Tableau 6. Série de 4 essais CNL, plaque rugueuse, amplitude variée en cours d’essai
Figure 8. Essais CNL plaque rugueuse, σ ncm 0 = 310 kPa, Δτ = 10 kPa, ID0 = 30 % (L…gauche), 90 %
9 9
(D…droite). LL, DL (η cm 0 = 0), LM, DM (η cm 0 ≈η pic / 2 ), LH ( η pic ) contractants ; DH( η pic ) dilatant.
10 10
313
Figure 9. Essais CNL plaque rugueuse, σ ncm 0 = 60, 120, et 310 kPa, Δτ = 10 kPa, ID0 = 30 %
(à gauche) et 90 % (à droite). η cm 0 ≈η pic / 2 . Tous contractants. Déplacement relatif tangentiel
fonction décroissante de la contrainte normale initiale.
Figure 10. essais CNL, plaque rugueuse. σ ncm 0 = 60 kPa, Δτ = 10 et 20 kPa, ID0 = 30 % (à
gauche) et 90 % (à droite). Effet de l’amplitude cyclique.
314
Les influences de la densité initiale et du niveau cyclique moyen sont visibles figure 8 :
dilatance seulement à densité et niveau cyclique moyen élevés. La figure 9, à cisaillement
cyclique moyen similaire (η cm 0 ≈η pic / 2 ), montre l’influence du niveau de contrainte normale
sur les déplacements relatifs normal et tangentiel. L’influence de l’amplitude cyclique se lit à
la figure 10, comparant des essais CNL à même contrainte cyclique moyenne (60 kPa), à
même amplitude cyclique (10 et 20 kPa), et à cisaillement cyclique moyen faible à fort. A
tous les cisaillements cycliques moyens, une amplitude cyclique plus élevée accroit le
déplacement relatif tangentiel. L’essai de cisaillement direct CNS typique est représenté
figure 11. Le comportement d’interface sol-structure (rugueuse) sur chemin CNS est toujours
contractant, conduisant à plus ou moins brève échéance, à l’état critique. Le nombre de
cycles conduisant à cet état est d’autant plus élevé que la densité est plus forte, que la rigidité
normale est plus faible, que le cisaillement cyclique moyen est plus faible, et que l’amplitude
cyclique est plus faible. Le niveau de contrainte est également très influent, si l’on compare
les nombres de cycles à la rupture aux figures 11 et 12, traduisant deux essais à mêmes
paramètres, sauf le niveau de contrainte initial (respectivement 310 kPa, 2035 cycles, et 100
kPa, 2 cycles).
La figure 13, présentant des essais cycliques à amplitudes successives variées (5000 petits
cycles suivis de 5000 grands cycles et l’inverse) montre l’influence prépondérante des grands
cycles.
Les essais CNL avec plaque lisse exhibent les mêmes tendances que dans le cas rugueux, avec
des spécificités remarquables (figure 14):
- le frottement limite est évidemment moindre,
- à nombre égal de cycles, et paramètres cycliques identiques, les déplacements relatifs
sont sensiblement plus élevés, qu’il s’agisse du déplacement relatif tangentiel ou
normal, ce qui est sans doute à mettre en relation avec une plus grande liberté de
mouvements des grains.
Figure 11. Essai CNS, plaque rugueuse, σ ncm 0 = 310 kPa, Δτ = 10 kPa, η cm 0 = 0,30, ID0 = 30
%. k = 1000 kPa/mm. Rupture (état critique) à 2035 cycles.
315
Figure 12. Essai CNS, plaque rugueuse, σ ncm 0 = 100 kPa, Δτ = 20 kPa, η cm 0 = 0,30, ID0 = 30
%. k = 1000 kPa/mm. Rupture (état critique) au bout de 2 cycles.
Figure 13. Comparaison d’essais cycliques CNL à amplitude variée, plaque rugueuse. σ ncm 0 =
120 kPa, η cm 0 = 0,30, à gauche ID0 = 30 %, à droite ID0 = 90 %. Δτ = 10 puis 20 kPa, et
l’inverse.
On notera que l’essai à densité initiale forte et à niveau cyclique moyen élevé développe de la
dilatance, même dans le cas de l’interface lisse.
316
Figure 14. Essais CNL, plaque lisse. Déplacements relatifs tangentiels et normaux, contrainte
normale initiale 120 kPa, densités faible (ID0 = 30 %) et forte (ID0 = 90 %), divers niveaux de
cisaillement cyclique moyen.
Figure 15. Essai CNS, plaque lisse, densité forte, contrainte normale initiale 100 kPa, k =
5000 kPa/mm. a) chemin de contraintes. b) Evolution de la contrainte normale en fonction du
déplacement relatif tangentiel. Rupture à 48 cycles.
Concernant les chemins CNS, la tendance reste la même que dans le cas de la plaque
rugueuse, contractance dans tous les cas, menant à la rupture (état critique) à plus ou moins
brève échéance. Mais le nombre de cycles nécessaire pour atteindre la rupture est
drastiquement plus faible que dans le cas rugueux (figure 15). La rupture intervient au bout de
317
48 cycles seulement dans le cas lisse, contre 1175 cycles dans le cas rugueux, les paramètres
cycliques étant rigoureusement les mêmes ( σ ncm 0 = 100kPa, η cm0 = 0,35, Δη = 0,2). Ceci dit,
les taux de cisaillement limites sont plus faibles pour la plaque lisse (η pic = 0,49 et η crit =
0,43 dans le cas lisse, contre 0,61 et 0,57 dans le cas rugueux).
Les analyses granulométriques opérées après cisaillement ont été jugées trop peu
discriminantes, même avec la technique laser, pour être exploitées, ainsi qu’en témoigne la
figure 16 (essai à densités initiales faible et forte, même contrainte normale cyclique moyenne
(310 kPa), même amplitude cyclique ( Δτ = 10 kPa), et cisaillements cycliques moyens
variés(η cm 0 = 0,30 à 0,60).
Figure 16 . Essais CNS, plaque rugueuse, σ ncm 0 = 310 kPa, Δτ = 10 kPa, η cm 0 variés (= 0,30
à 0,60), ID0 = 30 % et 90 %, k = 1000 kPa/mm. Analyses granulométriques avant et après
cisaillement..
Chapitre 3. Modélisation constitutive des essais de cisaillement direct
3.1. Un modèle phénoménologique « pré-intégré »
En situation 2-D, la modélisation constitutive du cisaillement direct d’interface sol-structure

consiste en la relation tangente entre réponse et sollicitation, ici entre les dérivées temporelles
des vecteurs déplacement relatif (entre sol et structure) et contrainte (agissant sur l’interface),
fonction du chemin suivi (c), et de paramètres de mémoire (mi), soit :
⎛ ⎡•⎤ ⎞
⎜ ⎢u ⎥ ⎟ ⎛ • ⎞
⎜ ⎣ ⎦ ⎟ = f (⎜ σ n ⎟, c, m ) (2)
⎜⎡ • ⎤⎟ ⎜ • ⎟ i
⎜ ⎢ w⎥ ⎟ τ
⎝ ⎠
⎝⎣ ⎦⎠
Le chemin (tangent) de cisaillement est mesurable, par exemple, par le rapport des
composantes de la sollicitation. Nous nous limiterons à une classe de chemins, dits à rigidité
normale imposée non négative (§ 2.1). Parmi ces chemins, certains servent à l’identification
du modèle, tandis que d’autres sont voués à la validation. Dès lors, si nous sommes en mesure
de donner une formulation analytique (évidemment approchée) d’un chemin fini à c constant,
et non plus de sa seule dérivée temporelle, nous considérerons cette formulation comme une
318
forme « pré-intégrée » de ce chemin sur un domaine fini de sollicitation, fonction des
conditions initiales. Dans notre cas, la mémoire se réduit à un seul paramètre, le poids
1
volumique sous contrainte, γ =γ ( ) , directement déductible du
d σn d 0 σ n 1 + [u ] / t
déplacement relatif normal [u ] , des conditions initiales, et de l’épaisseur d’interface t. Le

poids volumique sous contrainte peut outrepasser le poids volumique maximum
(conventionnel !).
3.2. Modélisation des essais monotones
Nous avons été en mesure de donner une formulation analytique de tous les essais monotones
(à contrainte normale constante et à rigidité normale imposée). Nous présentons dans la thèse
seulement des chemins d’identification, CNL et CNS, dans la mesure où, sauf chemins
rarement suivis, à savoir variation de contrainte normale en cours d’essai CNL ou CNS, ou de
rigidité normale imposée en cours d’essai CNS, les validations ne s’imposent pas. Au cas où
de tels chemins seraient nécessaires, le paramètre de mémoire cité précédemment, γ
d σn
serait utilisé pour enchaîner des séquences de sollicitations, sous la forme d’incréments finis,
formulés analytiquement. Les figures 17 et 18 présentent les paramètres clés des formulations
analytiques respectivement pour les chemins CNL et CNS. La pente de la courbe [u ] − [w] du
chemin CNL au-delà de [w]crit n’est pas classique, mais existe, surtout dans le cas d’une
interface sol friable-structure (sables carbonatés par exemple). De même, la pente de la courbe
σ n − [w] du chemin CNS existe principalement pour ce même type d’interface, la rupture des
grains intervenant, bien que moindre, même pour des sables siliceux. Il s’agit donc
d’identifier ces paramètres clés, avec leurs variations en fonction de l’indice de densité initiale
ID0, du niveau initial de contrainte σ n 0 , et de la rigidité normale imposée k (3 paramètres).
Nous avons repris, en les perfectionnant, les formulations de Hoteit (1990), Boutrif (1993), et
Garnica Anguas (1993). La figure 19 présente la modélisation d’essais CNL entre sable de
Fontainebleau et plaque rugueuse, à densités faible (30 %) à gauche, et forte (90 %) à droite.
Figure 17. Paramètres clés de formulation des chemins CNL. a) courbe τ − [w] . b) courbe
[u ] − [w] .
319
Figure 18. Paramètres clés de formulation des chemins CNS. a) courbe τ − [w] . b) courbe
σ n − [w] .
Figure 19. Comparaison entre essai et formulation analytique d’essais CNL, σ n 0 = 120 kPa,
densités faible (gauche) et forte (droite).
320
La figure 20 fournit la modélisation d’essais CNS entre sable de Fontainebleau et plaque
rugueuse, à densités faible (30 %) à gauche, et forte (90 %) à droite, pour diverses valeurs de
la rigidité normale imposée. Dans les 2 cas, figures 19 et 20, la contractance à densité faible et
la dilatance à densité forte, lors de grands déplacements relatifs tangentiels, sont bien simulés.
Figure 20. Comparaison entre essais et formulation analytique d’essais CNS, σ n 0 = 100 kPa,
densités faible (gauche) et forte (droite). k = 1000, 2000, et 5000 kPa/mm.
3.3. Modélisation des essais cycliques CNL d’identification
Rappelons les grands principes du modèle élasto-visco-plastique de Perzyna (1966) adapté au

pseudo-fluage cyclique : Au cours des cycles, le vecteur incrément de déplacement relatif sol-
structure est la somme d’une partie élastique indicée « e », et d’une partie cyclique moyenne
d [u ] = d [u ]e + d [u ]cm (3)
indicée « cm ». La première partie, élastique, est classiquement écrite ( De matrice élastique) :
d [u ]e =
−1
D e
dσ (4)
321
tandis que le vecteur incrément de déplacement relatif visco-plastique (ici cyclique moyen), a
pour expression ( Γ paramètre scalaire de fluidité, F frontière entre comportements non
visqueux et visqueux, Φ noyau visqueux (0 à l’intérieur, 1 à l’extérieur de F), G (σ ) potentiel
visco-plastique, dont les dérivées partielles définissent la direction de l’écoulement visqueux,
dN incrément de nombre de cycles):
∂G
d [u ]cm = ΓΦ ( F ) dN (5)
∂σ
Dans notre cas, le noyau est vide, ce qui signifie que le pseudo-fluage existe quel que soit le
niveau des paramètres cycliques. Par ailleurs le vecteur [u ]cm est directement exprimé
⎛ [u ] ⎞
analytiquement, par ses composantes ⎜⎜ cm ⎟⎟ , fonctions de N, qui traduisent en outre, de
⎝ [w]cm ⎠
manière naturelle, la direction d’écoulement pseudo-visqueux. Le déplacement relatif
irréversible [u ]cm , déclaré pseudo-visqueux, est en fait une accumulation de déplacements
relatifs plastiques.
3.3.1. Formulation analytique des chemins clés CNL de cisaillement cyclique
La base de notre interprétation est classique (figure 21), ce qui nous a conduit à
détecter/identifier, les 3 sortes de pseudo-fluage (primaire, secondaire, et tertiaire), la vitesse
Figure 21. Le fluage, tel qu’il est classiquement décrit.
∂ ∂
de pseudo-fluage s’entendant comme au lieu de (t = temps en fluage). Les paramètres
∂N ∂t
de nos essais sont (§ 2.6.1) ID0, σ ncm 0 , η cm 0 , Δη 0 , et N. A l’examen attentif des essais, le plus
souvent menés jusqu’à 10000 cycles, il est apparu les caractéristiques générales
suivantes pour les composantes [ w]cm et [u ]cm :
- La composante [ w]cm est identiquement nulle pour des cycles centrés (η cm 0 = 0 ).
- Le pseudo-fluage primaire existe pour les 2 composantes ( [ w]cm et [u ]cm ), et pour
tous les essais (noyau visqueux vide au sens de Perzyna).
322
- Le pseudo-fluage secondaire existe pour les 2 composantes dans les essais à valeurs
modérées du cisaillement cyclique moyen réduit η cm 0 . En outre ces essais (figure 22)
sont toujours contractants ( [u ]cm < 0 ).
- Concernant le déplacement relatif tangentiel cyclique moyen [ w]cm , le pseudo-fluage
tertiaire existe aux valeurs élevées de η cm 0 .
- Concernant le déplacement relatif normal cyclique moyen [u ]cm aux valeurs élevées
de η cm 0 , on peut constater (figure 22) une dilatance (ID0 élevé), ou une contractance
(ID0 faible), aboutissant à une valeur limite [u ]cm l (en relation avec la densité critique
sous contrainte γ , avec t épaisseur d’interface).
d crit σ n
1
γ =γ ( ) (6)
d crit σ n d0 σn 1 + [u ]cm l / t
Figure 22. Les 3 « pôles » du comportement d’interface sable-structure, relativement au

déplacement relatif normal cyclique moyen [u]cm , en relation avec les résistances de pic et
critique, et avec l’angle « caractéristique » (Luong, 1980).
- Le niveau de contrainte normale cyclique moyenne σ ncm 0 agit directement sur les
vitesses de pseudo-fluage cyclique, un peu à la manière d’un module (A contrainte
faible vitesse élevée, et à contrainte élevée vitesse faible).
- L’amplitude cyclique réduite Δη 0 intervient comme un amplificateur, étant entendu
que le pseudo-fluage cyclique s’éteint à amplitude cyclique nulle.
L’évolution de [u]cm est ainsi définie pour les pôles (nœuds en langage éléments finis):
- a) Cisaillement réduit modéré, c'est-à-dire η cm 0 = η * = η car / 2 par:
γ = γ d* (Δη ) n constantes fonctions de σ ncm et de ID0 ; Nf = 10000 cycles (7)
d max σ n N f
N η max
Evolution hyperbolique f ( N ) = , N 50% fonction de σ ncm , ID0, (8)
N 50% + N η pic
- b) Cisaillement réduit élevé, c'est-à-dire η cm 0 > η car
γ fonction de σ ncm (§ 2.5.1). Ce poids volumique constitue une asymptote/N
d crit σ n
323
Evolution hyperbolique comme en a)
- c) en outre, l’évolution de [u]cm pour η cm 0 = 0 est lièe à a) par le paramètre a :
γ = aγ d* (Δη ) n avec 0 < a < 1 , a fonction de σ ncm et de ID0 définir Nf (9)
d max σ n N f η cm = 0
L’évolution de [ w]cm ne requiert de formulation que dans les cas a et b :

- a) Cisaillement réduit modéré : Il s’agit de pseudo-fluages cycliques primaire et
secondaire. Nous avons défini (figure 23) une asymptote (pente a, ordonnée à l’origine
b dans le plan ( [ w]cm -N), et une loi d’évolution hyperbolique y avec N 50% (C sur
figure 23) double du précédent.
Figure 23. Définition de la fonction [ w]cm ( N ) , η cm 0 modéré, pseudo-fluages cycliques

primaire et secondaire, courbe y.
- b) Cisaillement réduit élevé, c'est-à-dire η cm 0 > η car : Le pseudo-fluage cyclique

secondaire est pratiquement absent. Nous avons défini une pente initiale ( y 0' ), une
Figure 24. Définition de la fonction [ w]cm ( N ) , η cm 0 élevé, pseudo-fluages cycliques primaire

et tertiaire, secondaire pratiquement inexistant, courbe y interpolée entre y1 et y2.
324
asymptote (a et b), ainsi qu’une interpolation exponentielle entre ces 2 extrêmes (figure 24).
Les paramètres y 0' , a et b (paramètre courant p) sont fonctions de η cm 0 , Δη (figure 25), et
σ ncm . Pour Δη < Δη max , interpolations linéaire ou quadratique.
Figure 25. Limites et allure du paramètre p (= y 0' /a / b), pseudo-fluages primaire et tertiaire,
cas η cm 0 élevé.
3.3.2. Interpolation constitutive
On se réfère ici au langage éléments finis. L’interpolation constitutive consiste en une

interpolation entre les comportements clés (« nœuds » correspondant à η cm 0 = η * = η car / 2 ,
η cm 0 ≈ η crit , accessoirement 0 < η cm 0 < η * ) définis précédemment (§ 3.3.1).
Figure 26. Les fonctions d’interpolation W et W , pour [u ]cm ( a ≠ 0 ), cas dense

c corr d corr
( I D 0 = 90% ) et niveau de contrainte élevé ( σ ncm 0 = 310 kPa ).
325
Nous avons construit des interpolations constitutives voisines pour [u ]cm et [w]cm , la
différence provenant du paramètre a , relatif aux chemins η cm 0 = 0 , avec a = 0 pour [u ]cm et
0 < a < 1 pour [w]cm . Les fonctions d’interpolation sont relatives à un niveau de contrainte
(car η pic en dépend), et à un niveau de densité (pour la même raison). Deux fonctions
d’interpolation dérivant du type hermitien sont utilisées : Wc (indice c comme contractant)
centrée sur ± η * , et Wd (indice d comme dilatant) centrée sur ± η pic . La difficulté a été de
générer une seule fonction analytique définie sur le domaine − η pic ,+η pic pour chacune
d’entre elles. D’autant plus qu’il s’agit d’une vraie interpolation sur les fractions de domaine
− η pic ,−η * et η * ,+η pic (alors Wc + Wd = 1 ), tandis que sur la fraction de domaine
− η * ,+η * , Wd n’intervient plus (alors Wd ≡ 0 ), et Wc varie entre a et 1. Nous avons utilisé
une fonction de correction très voisine d’un « créneau », analytique elle aussi, pour parvenir à
ce résultat. Les fonctions d’interpolation utilisées sont finalement W et W . La figure
c corr d corr
26 donne une représentation de ces fonctions dans le cas dense et pour le niveau de contrainte
σ ncm 0 = 310 kPa .
Seuls les essais avec plaque rugueuse sont présentés ici. Les figures 27 ( σ ncm 0 = 120 kPa ;
η cm 0 faible, moyen, élevé ; faible amplitude cyclique ; I D 0 = 30%) et 28 ( σ ncm 0 = 310 kPa ;
η cm 0 faible, moyen, élevé ; faible amplitude cyclique ; I D 0 = 90%) montrent des résultats
expérimentaux comparés aux courbes analytiques d’identification. On notera la valeur limite
de [u ]cm correspondant au poids volumique critique d’interface, pour η cm 0 élevé.
Figure 27. Essais CNL et identification analytique : σ ncm 0 = 120 kPa ; η cm 0 faible (= 0),
moyen ( ≈ η pic / 2 ), élevé ( ≈ 0.8 × η pic ); faible amplitude cyclique ( Δτ = 10 kPa); I D 0 = 30%.
326
Figure 28. Essais CNL et identification analytique : σ ncm 0 = 310 kPa ; η cm 0 faible (= 0),
moyen ( ≈ η pic / 2 ), élevé ( ≈ 0.8 × η pic ); faible amplitude cyclique ( Δτ = 10 kPa); I D 0 = 90%.
Figure 29. Généralisation analytique à un chemin CNL à densité moyenne ( I D 0 = 60%) par
interpolation linéaire : σ ncm 0 = 310 kPa ; η cm 0 faible (= 0), moyen ( ≈ η pic / 2 ),
élevé ( ≈ 0.8 × η pic ); faible amplitude cyclique ( Δτ = 10 kPa).
Les formulations analytiques des essais CNL d’identification permettent un premier niveau de
généralisation qui pourrait servir à la validation si d’autres essais existaient. On peut ainsi
étendre ces résultats par interpolation linéaire sur la densité (figure 29 pour I D 0 = 60%),
quadratique sur le niveau de contrainte normale cyclique moyenne (figure 30 pour σ ncm 0 = 200
kPa).
327
Figure 30. Généralisation analytique à un chemin CNL à niveau moyen de contrainte cyclique
moyenne ( σ ncm 0 = 200 kPa ) par interpolation quadratique : η cm 0 moyen ( ≈ η pic / 2 );
amplitude cyclique moyenne ( Δτ = 20 kPa). I D 0 = 30% et 90 %.
3.4. Modélisation d’essais CNL à amplitude cyclique variée
Comme nous l’avons indiqué précédemment, les formulations analytiques sur chemin CNL
⎛ [u ] ⎞
fournissent ⎜⎜ cm ⎟⎟ fonction des paramètres ID0, σ ncm 0 , η cm 0 , Δη 0 , et N. Comme la mémoire
⎝ [w]cm ⎠
du comportement d’interface est représentée par γ directement relié à [u ]cm , l’évolution de
d σn
cette mémoire est immédiatement disponible sur de tels chemins. Ce n’est plus le cas lorsque
le chemin CNL choisi inclut un changement de l’un des paramètres σ ncm 0 , η cm 0 , Δη 0 (par
exemple plusieurs séries de cycles d’amplitude variée ou de niveaux moyens variés). Il s’agit
alors de passer de la formulation analytique « précédente », à la formulation analytique
« suivante », le lien entre celles-ci étant le paramètre de mémoire, démarche que nous
expliquons ci-dessous. Néanmoins, notre analyse utilise des incréments analytiques finis ou
pré-intégrés.
3.4.1. Nombre équivalent de cycles. Validation sur essais CNL à amplitudes variées
La figure 31 donne le principe de la recherche itérative du passage entre les 2 formulations,

qui fait intervenir le « nombre équivalent de cycles » N eq de la série dite « suivante »,
328
équivalent à l’évolution au cours des séries dites « précédentes » de cycles. Autrement dit,
étant donné l’état atteint par l’interface au terme de la série (i-1), soit :
ℑi −1 = {ID0, σ ncm 0 , η cm 0 , Δη i −1 , N i −1 , [u ]cm i −1 , [w]cm i −1 } (10)
Nous recherchons l’état de début de la série de cycles i, à savoir :
ℑi = {ID0, σ ncm 0 , η cm 0 , Δη i , N ≠ N i −1 , [u ]cm i = [u ]cm i −1 , [w]cm i } (11)
eq i
Figure 31. Principe d’évaluation du nombre équivalent de cycles, entre 2 séries successives de
cycles de paramètres différents.
Il n’y a pas forcément coïncidence entre [w]cm i et [w]cm i −1 , ce que nous considérons comme
secondaire. La figure 32 donne une comparaison expérience modélisation pour 2 séries
successives de 5000 cycles, à σ ncm 0 = 120 kPa ), η cm 0 moyen ( ≈ η pic / 2 ); amplitude
cyclique moyenne ( Δτ = 10 puis 20 kPa). I D 0 = 30%.
Figure 32. Comparaison expérience modélisation pour 2 séries successives de 5000 cycles, à
σ ncm 0 = 120 kPa , η cm 0 moyen ( ≈ η pic / 2 ), amplitude cyclique moyenne ( Δτ = 10 puis 20
kPa). I D 0 = 30%.
329
3.4.2. Modélisation de chemins CNL possibles de type tempête
A titre d’exemple, des séries de cycles de type tempête sont modélisées figure 33, avec les
paramètres σ ncm 0 = 200 kPa , η cm 0 = 0,34, I D 0 = 90%, 2 séquences de chargement (S1 et S2)
de chacune 5 séries d’amplitudes cycliques différentes.
Figure 33. Deux séquences de chargement (S1 et S2) de chacune 5 séries d’amplitudes
cycliques différentes, paramètres σ ncm 0 = 200 kPa , η cm 0 = 0,34, I D 0 = 90%.
3.4. Modélisation des essais cycliques CNS
Dans le cas des chemins cycliques CNS, une complexité supplémentaire provient des
variations possibles de la contrainte normale cyclique moyenne σ ncm , ,et par conséquent du
cisaillement cyclique moyen η cm et de l’amplitude cyclique réduite Δη . Un incrément de
chemin CNS se calcule donc comme la combinaison (on parle de « superposition ») d’un
incrément CNL et d’un incrément pseudo-oedométrique (figure 34). Par ailleurs, le tampon
oedométrique situé au dessus de l’interface, qui était considéré comme inerte pendant le
cisaillement sur chemin CNL, devient actif, sa compressibilité ( K n ) agissant en série avec la
rigidité normale externe imposée k (figure 35) selon (12).
Δσ
Δ[u ]cm Δ[u ]cm
k * Kn
Δσ = =
ncm CNS
et (12)
Kn + k
CNL CNS
ncm CNS k
330
Figure 34. Incrément de chemin CNS = incrément de chemin CNL + incrément de chemin
pseudo-oedométrique.
Figure 35. Correction du déplacement relatif normal d’interface, due à la compressibilité du

tampon pseudo-oedomètrique sur chemin incrémental CNS.
3.4.1. Intégration par « incréments finis, pré-intégrés »
Nous avons poursuivi notre approche par incréments finis analytiques (donc pré-intégrés) de
taille maxima compatible avec la stabilité d’intégration. La figure 36 montre l’effet du pas
d’intégration, qui a été contrôlé dans tous nos résultats. Cette figure est relative à un
échantillon initialement dense, pour lequel une intégration par pas de 200 cycles présente des
331
oscillations inacceptables, tandis qu’un pas de 10 cycles donne satisfaction. Dans le cas lâche,
non représenté ici, une bonne stabilité d’intégration a conduit à des pas très faibles, parfois
inférieurs à 1 cycle.
Figure 36. Caractère stable ou oscillatoire de l’intégration sur chemin CNS, selon la taille du
pas d’intégration ( σ ncm 0 = 100 kPa , η cm 0 = 0.50, Δη 0 = 0.1, I D 0 = 90%).
L’intégration sur chemins CNS est d’autant plus aisée que plus le niveau de contrainte est plus
élevé, que la densité est plus forte, et que le cisaillement cyclique réduit est plus faible. Les
figures 37 et 38 montrent des cas bien simulés. Une information très intéressante et concise
Figure 37. Chemin CNS , plaque rugueuse, ,essai et modélisation ; k = 1000 kPa/mm, σ ncm 0 =
100 kPa , η cm 0 = 0.35, Δη 0 = 0.1, I D 0 = 90%.
332
Figure 38. Chemin CNS , plaque rugueuse, essai et modélisation ; k = 1000 kPa/mm, σ ncm 0 =
310 kPa , η cm 0 = 0.30, Δη 0 = 0.032, I D 0 = 30%.
Figure 39. Ncrit sur chemin CNS , plaque rugueuse, I D 0 = 30%, essai et modélisation :
Comparaison et modèle d’évolution.
333
consiste en la nombre de cycles menant à la rupture, ou à l’état critique, Ncrit, sur chemin CNS
qui, rappelons le, sont toujours contractant. Dans l’espace des contraintes, et selon nos essais,
ces chemins sont horizontaux, c'est-à-dire à contrainte de cisaillement cyclique moyenne
constante. Les figures 39 et 40 permettent une comparaison entre les cas lâche et dense, pour
plaque rugueuse. Un résultat schématique : les valeurs de sont 100 fois plus élevées entre le
cas dense et le cas lâche.
Figure 40. Ncrit sur chemin CNS , plaque rugueuse, I D 0 = 90%, essai et modélisation :
Comparaison et modèle d’évolution.
Chapitre 4. Modélisation par la méthode des éléments finis
4.1. Introduction
Nous avons utilisé dans cette partie de la thèse le logiciel éléments finis PLAXIS, dédié à la
géomécanique. Le modèle constitutif choisi est le plus simple, Mohr-Coulomb à 5 paramètres
(E, ν, ,c, ϕ, ψ). En ce qui nous concerne, un des intérêts originaux du modèle structural
PLAXIS est le traitement efficace des interfaces entre matériaux, en particulier des interfaces
sol-structure.
Nous concevons la modélisation numérique guidée par le canevas suivant :
- Il s’agit d’abord d’identifier les mécanismes que nous supposons actifs dans
l’expérience à modéliser.
- Puis il convient de classer ces mécanismes dans l’ordre hiérarchique d’importance que
nous imaginons.
- Ensuite, nous devons implémenter ces mécanismes dans le modèle structural, autant
que faire se peut, en privilégiant d’abord les plus importants.
334
- Enfin, il n’est pas inutile de lister les mécanismes négligés, en se demandant
pourquoi ils l’ont été.
Nous avons tenté de modéliser, sur la base de ces idées, l’essai de cisaillement direct
monotone, ainsi que des sollicitations cycliques axiales à grand nombre de cycles sur pieux
modèles centrifugés.
4.2. Modélisation 2-D de l’essai de cisaillement direct sol-structure monotone à la boîte

de cisaillement
Nous avons souhaité examiner, par modélisation numérique, les hétérogénéités si souvent
décriées, de la boîte de cisaillement. Les essais d’Hoteit (1990) réalisés à l’aide d’un boîte
carrée munie d’une plaque latérale de verre, et analysés par stéréophotogrammétrie, ont monté
que la boîte fonctionnait selon 2 couches, l’une d’interface de faible épaisseur, et l’autre
comme un tampon oedométrique, ce qui a motivé notre interprétation (§ 2.2.5).
Conformément au canevas ci-dessus, les aspects et mécanismes présents lors du cisaillement

direct à la boîte sont, dans l’ordre hiérarchique :
- le caractère 3D de l’essai
- le frottement sol-plaque rugueuse et sol-boîte
- le chemin de cisaillement direct
- le frottement piston-boîte
Le caractère 3D de l’essai ne peut être simulé en 2D ; nous avons simulé l’essai en
déformation plane, avec des éléments de volume T15, et des éléments d’interface J10. Les
caractéristiques mécaniques du sable, de la plaque rugueuse, de la boîte et du piston sont
indiquées au tableau 7. Des interfaces existent aux contacts sable plaque, sable boîte, et boîte
plaque. La géométrie du modèle, le maillage, la sollicitation (verticalement contrainte
normale sur le sable, horizontalement déplacement imposé de la boîte contenant le sable), et
les points de contrôle (A à F) sont rapportés figure 41 pour le chemin CNL. L’hétérogénéité
en terme de contrainte de cisaillement se manifeste figure 42. On constate une absence de pic
de cisaillement (figure 43) et corrélativement, un déplacement relatif normal sous-évalué
(C’est le plus grave défaut que l’on peut reprocher à cette modélisation). Le chemin CNS est
modélisé avec la même géométrie de base et les mêmes caractéristiques mécaniques que le
chemin CNL, mais la sollicitation supérieure est à déplacement imposé sur un piston
compressible (figure 44), ce qui simule la rigidité normale imposée. Les principaux aspects du
Tableau 7. Caractéristiques mécaniques de volumes et d’interfaces utilisées pour modéliser

l’essai de cisaillement direct monotone.
335
Figure 41. Modélisation de l’essai de cisaillement direct sur chemin CNL.
Figure 42. Hétérogénéité à l’interface sable plaque, du cisaillement simulé sur chemin CNL.
Figure 43. Modélisation de l’essai de cisaillement direct CNL (valeur globale moyenne de la
contrainte de cisaillement). a) densité faible. b) densité forte.
336
cisaillement direct CNS sont rendus (figure 45), à savoir la dilatance à densité élevée, la
contractance à densité faible, et ce quel que soit le niveau de contrainte normale (60 à 310
kPa). Par contre, le contraste pic palier et le retour vers l’origine le long de la ligne critique ne
sont pas simulés, ce qui n’est pas inattendu avec un modèle constitutif de Mohr Coulomb.
Figure 44. Modélisation de l’essai de cisaillement direct sur chemin CNS.
Figure 45. Modélisation de l’essai de cisaillement direct CNS (valeur globale moyenne des
contraintes normale et de cisaillement). A gauche densité faible, à droite densité forte.
337
4.3. Modélisation des sollicitations cycliques à grand nombre de cycles
Nous avons pu procéder à ces modélisations grâce aux essais de pieux modèles centrifugés
(23g) réalisés par l’IFSTTAR, et mis à disposition de la communauté SOLCYP. La figure 46
présente la nacelle de la centrifugeuse, et 8 pieux d’essai régulièrement répartis, ainsi qu’une
vue du dispositif embarqué de chargement. Le tableau 8 donne les rapports de similitude pour
Figure 46. Dispositif expérimental d’essais de pieux modèles, échelle 1/23, à l’IFSTTAR,
Nantes.
les diverses grandeurs physiques en jeu. Rappelons que les contraintes sont conservées entre
le prototype et le modèle réduit. Les pieux modèles sont en aluminium, de diamètre 18 mm,
Tableau 8. Rapports de similitude pour les diverses grandeurs physiques en centrifugeuse.
de longueur 590 mm (dont 30 mm en tête émergeant du sable de Fontainebleau de la nacelle,

de masse 0,405 kg, et rugueux sur toute leur surface latérale. Les frontières du modèle sont, à
la profondeur 690 mm pour la base, et au rayon 240 mm pour la frontière latérale. Selon les
338
données IFSTTAR, ce sable de Fontainebleau possède un angle de frottement de l’ordre de
40°, un angle de dilatance de 14°, et les caractéristiques d’interface mesurée à la boîte de
cisaillement sont légèrement inférieures. Par ailleurs, les essais de pieu modèles sous
sollicitations axiales monotones, en traction et compression ont donné les résultats résumés au
tableau 9.
Tableau 9. Résultats d’essais de pieux modèles centrifugés (23g) sous sollicitations axiales
monotones en traction et compression.
Revenant au canevas exposé § 4.1, nous avons fait les choix suivants, en privilégiant la
simplicité et la rusticité, compte tenu que nous avons placé la plus grande partie de notre
énergie et de notre temps de travail dans nos essais de cisaillement direct cycliques. Il s’agit
donc ici d’exposer une méthodologie, et de montrer sa faisabilité, étant entendu qu’il y aurait
lieu de raffiner ultérieurement cette approche, à de nombreux points de vue, et notamment de
rendre automatique le dialogue EXCEL-PLAXIS, par fichiers, ce qui est actuellement tout à
la charge de l’opérateur :
- modélisation en axisymétrie, éléments de volume T15, et d’interface J10.
- pieu modélisé en éléments de volume, et sollicité par la force en tête.
- Sol (sable) modélisé selon 3 couches horizontales pour la sollicitation monotone, 6
couches horizontales pour la sollicitation cyclique
- interfaces sol pieu le long du fût et en pointe.
- modèle constitutif élasto-plastique de Mohr Coulomb (presque) classique pour le
sable. Les aspects cycliques ne sont pas pris en compte dans le sol, car vu la
géométrie, le cisaillement décroît très rapidement dans la direction radiale à partir du
fût (en 1/r), ce qui limite naturellement ces effets.
- poids volumique du sol adapté, tenant compte de l’accélération imposée par la
centrifugeuse, proportionnelle au rayon à partir de l’axe de rotation, et valant 23g à
240 mm de profondeur dans la nacelle.
- Module d’Young du sol adapté, linéaire par rapport à la profondeur dans la nacelle
(figure 47), et approximation selon les moindres carrés par rapport à la loi de Kondner
n
⎛ σ ⎞
( E = E ref ⎜ i ⎟ ). Ici, n = 0.7, σ ref = 100 kPa, E ref = 4500 kPa et σ i contrainte
⎜σ ⎟
⎝ ref ⎠
moyenne, en kPa, résultant des contraintes pseudo-géostatiques (23g) et K0 = 0.7 (car
plusieurs cycles de centrifugation avant essai proprement dit).
- A un stade donné de la sollicitation en tête de pieu, PLAXIS fournit une
approximation supposée acceptable des chemins cycliques locaux d’interface, par un
calcul pas à pas d’un cycle entier. Par contre, le modèle constitutif de PLAXIS est
insuffisant pour modéliser la contractance ou la dilatance cyclique. Donc l’ensemble
des cycles imposés en tête de pieu est divisé en plusieurs séries, pour lesquelles on
339
calcule le premier cycle de la série à l’aide de PLAXIS, ce qui donne les informations
locales pour la dite série.
- A l’interface sol-pieu le long du fût, on souhaite introduire les chemins de cisaillement
2G
direct cycliques CNS (localement k = , G module de cisaillement, R rayon du
R
pieu) qui nous l’avons vu, sont toujours contractants. Introduire notre modèle de
cisaillement cyclique comme modèle utilisateur a été impossible vu le temps imparti.
Nous avons choisi une solution approchée, et simplifiée. Nous avons ainsi décidé
d’utiliser les données de Ncrit extraites de nos essais (Rappelons –figures 39 et 40, §
3.4.2- que Ncrit représente pour un chemin CNS comme ceux de nos essais, le nombre
de cycles nécessaire pour parvenir à l’état critique, à τ cm et Δτ fixés). Le nombre de
cycles N, entre 0 et Ncrit , donne accès à la diminution de contrainte normale cyclique
moyenne sur chemin CNS, soit Δσ ncm , et accessoirement Δη cm . La loi d’évolution en
fonction de N est supposée, compte tenu de notre connaissance des chemins CNS.
- En toute rigueur, il conviendrait de générer Δσ ncm aux points d’intégration de
l’interface, ce qui n’est pas possible actuellement dans PLAXIS. Mais il existe un
moyen pour générer Δσ ncm par cluster (ensemble d’éléments), en y imposant une
variation (diminution) locale de volume ( ε V ). Nous avons donc créé dans PLAXIS un
ensemble de clusters de très faible épaisseur, contigus au pieu, et de propriétés
identiques à celles du sol, qui remplissent ce rôle. Ils seront nommés dans la suite
« clusters de contrôle » et indicés i, de 1 à nc.
- Les 2 modèles de pieux chargés cycliquement en traction et en compression utilisent
exactement les mêmes données géométriques et mécaniques, à une différence près : en
compression cyclique, le niveau des modules d’Young de la couche de sable située
sous la pointe a été majoré pour tenir compte de la contrainte moyenne élevée dans
cette zone.
- En traction comme en compression, le calcul est agencé selon les phases suivantes,
impliquant un dialogue EXCEL-PLAXIS :
0. Initialisation géostatique (1g) dans le module INPUT, selon la procédure K0,

pour la couche de sable située sous la pointe.
1. Initialisation géostatique (1g) dans le module CALCULATION, pour tout le
sol et pour le pieu métallique.
2. Suite des phases, 2 à 5, dans le module CALCULATION. Et tout d’abord
centrifugation du modèle (23g).
3. Début du traitement de la série (S) de cycles, comportant Ns cycles

identiques en tête. Application de la charge nominale moyenne en tête pour la
série (s) de cycles actuellement traitée.
4. Chargement en tête jusqu’à la charge maximum, puis minimum, puis
nominale moyenne, et enfin à nouveau maximum1.
5. Extraction, depuis les nc « clusters de contrôle », des paramètres locaux des
cycles, stockage dans EXCEL.
6. Calcul, dans EXCEL, des nc chemins CNS correspondants aux paramètres

extraits en phase 5. Ou, solution simplifiée, calcul des nc valeurs Ncrit
1
Ceci afin de connaître les limites locales liées à un cycle en tête, car la première montée en charge suivie de la
première descente en charge sont très influencées par l’état initial en fin de centrifugation sous 23g.
340
correspondantes. Calcul, dans chaque cluster de contrôle, des Δσ ncm
(contractance, donc <0) correspondant aux Ns cycles de cette série (S).
7. Création, dans PLAXIS CALCULATION, de nc cas de charge,
correspondant chacun à ε V unitaire (contractant = -1) dans l’un des clusters, =0
dans tous les autres.
8. Extraction, depuis les nc « clusters de contrôle », des Δσ ncm correspondant à
chaque cas de charge, stockage dans EXCEL.
9. Calcul, dans EXCEL, de la combinaison linéaire des nc cas de charge de (7)
permettant d’obtenir le champ des Δσ ncm définis en (6).
10. Création, dans PLAXIS CALCULATION, de la phase de pseudo-fluage

cyclique modélisant les Ns cycles de la série (S), à partir de la charge
nominale moyenne en tête atteinte en (4). Il s’agit de la combinaison linéaire
définie dans (9) des champs de ε V définis en (7).
11. Si le résultat en termes de champ de Δσ ncm définis en (6) n’est pas tout à
fait atteint, itération supplémentaire à partir du résultat atteint en (10)2.
12. Passage à la série suivante de cycles en tête : retour en (3), sinon fin des
calculs.
Figure 47. Module d’Young fonction de la profondeur dans la nacelle, approximation de la loi
de Kondner.
4.3.1. Pieu modèle centrifugé en traction, monotone et cyclique
Nous avons modélisé l’essai monotone (tableau 9) et l’essai cyclique en traction CO7TO5. La
figure 48 donne les résultats de l’application de l’accélération 23g sur la modèle. On note un
frottement négatif très prononcé, comme dans une installation de pieu par fonçage ou battage,
ce qui est dû au contraste de modules entre sol et pieu lors de ce type de chargement.
2
En effet, dans PLAXIS (cf Reference Manual, § 4.7.6), pour des raisons qu’il serait trop long d’expliquer ici,
une partie seulement de la variation de volume que l’on souhaite, est réellement imposée (de l’ordre de 50 %).
341
Figure 48. Maillage du modèle CO7TO5, application de l’accélération 23g sur le modèle :
déplacements (amplifié 20 fois), contraintes agissant sur le fût, déplacement vertical du pieu.
La figure 49a représente le chargement du pieu en traction monotone jusqu’à 4 kN, après
application des 23g. On note un déplacement en tête de 1.20 mm, voisin de celui mesuré en
centrifugeuse (1.30 mm, tableau 9). La figure 49b donne la répartition entre effort en tête et
force de pointe, à différents stades. La figure 50 fournit les distributions de contraintes
normale et de cisaillement sur le fût, aux mêmes stades de chargement. L’intense frottement
(a) (b)
Figure 49. Chargement monotone du pieu modèle en traction à 4 kN. a) charge-déplacement
en tête. b) charge en pointe versus charge en tête ; pointe/tête ≈ 0 %..
342
négatif initial est confirmé. Le calcul cyclique est effectué avec 6 couches horizontales, dont 5
contigües au pieu (figure 51a). On réalise le parcours pas à pas du premier cycle (figure 51b),
puis le pseudo-fluage cyclique au cours des 2000 cycles suivants. La figure 52 donne les
chemins locaux de contraintes à diverses profondeurs, le long du pieu. On doit noter que ces
Figure 50. Chargement monotone du pieu modèle en traction . a) distribution des contraintes
normales sur le fût. b) distribution des contraintes de cisaillement.
(a) (b)
Figure 51. Chargement cyclique du pieu modèle en traction, modélisation de l’essai CO7TO5.
a) maillage en 6 couches horizontales. b) charge en tête versus déplacement en tête, 1er cycle
entre 1 et 3 kN (en bleu), comparaison au chargement monotone.
chemins sont d’autant plus amples en termes de cisaillement, que l’on se déplace de la tête
vers la pointe du pieu. Le tableau 10 résume les paramètres cycliques de ces cycles, et la
rigidité normale locale du massif de sable. La figure 53 donne les chutes de contraintes
normales qui en découlent, en fonction du nombre de cycles, selon l’approche déduite de Ncrit.
Finalement, la figure 54 donne le déplacement cyclique moyen en tête de pieu résultant du
pseudo-fluage cyclique au bout de 2000 cycles, tandis que la figure 55 compare les
distributions de contraintes le long du pieu en traction, initialement (23g), au bout du premier
343
cycle, et au bout de 2000 cycles. La zone près de la pointe a perdu toute résistance. Le
déplacement cyclique moyen atteint au bout de 2000 cycles est de l’ordre de 1.54 mm.
Figure 52. Chemins de contraintes au cours du premier cycle, sur le fût du pieu modèle en
traction, modélisation de l’essai CO7TO5. layer 1 près de la surface libre de la nacelle, layer 5
près de la pointe du pieu.
Figure 53. Pieu modèle en traction. Chutes de contraintes normales sur chemins CNS le long
du pieu, en fonction du nombre de cycles, découlant des paramètres cycliques locaux, selon
l’approche déduite de Ncrit.
344
Tableau 10. Paramètres cycliques des cycles locaux le long du pieu en traction, au cours du
premier cycle de force en tête, et rigidité normale locale du massif de sable.
Figure 54. Pieu modèle en traction. Modélisation de l’essai CO7TO5 ; Force cyclique
moyenne-déplacement cyclique moyen en tête de pieu résultant du pseudo-fluage cyclique au
bout de 2000 cycles.
Figure 55. Modélisation de l’essai CO7TO5, traction cyclique. Distribution de contraintes le

long du pieu initialement (23g), au bout du premier cycle, et au bout de 2000 cycles.
345
4.3.2. Pieu modèle centrifugé en compression, monotone et cyclique
Nous avons modélisé l’essai monotone (tableau 9) et l’essai cyclique en compression

CO4TO7. Les résultats de l’application de l’accélération 23g sur la modèle sont évidemment
identiques aux précédents, au déplacement du pieu, en tête notamment près, puisque le
module d’Young de la couche située sous la pointe a été renforcé, compte tenu du niveau
élevé de contrainte dans cette zone. On note de toutes manières un frottement négatif initial
très prononcé, comme dans une installation de pieu par fonçage ou battage. La relation charge
-tassement sous 6.1 kN est présentée figure 56, tandis que la distribution de contraintes fait
l’objet de la figure 57. Le déplacement modélisé dû au chargement est de 1.51 mm, supérieur
Figure 56. Modélisation de l’essai de compression monotone du pieu modèle centrifugé,

courbe charge-tassement jusqu’à 6.1 kN.
Figure 57. Modélisation de l’essai de compression monotone du pieu modèle centrifugé,

distribution des contraintes normale et de cisaillement le long du fût.
346
à celui qui a été mesuré (tableau 9). La figure 58 compare les distributions de contraintes sous
chargements monotones en traction (4 kN) et en compression (6.1 kN). On note que les
contraintes normales sont voisines, excepté près de la pointe du pieu. Elle sont majorée en
moyenne de 70 % près de la point du pieu, entre les cas traction et compression. La figure 59
donne la répartition entre forces en tête et de pointe pour le pieu en compression, depuis
l’application de la gravité (1g).
Figure 58. Distributions des contraintes normale et de cisaillement le long du pieu modélisé
sous chargements monotones en traction (4 kN) et en compression (6.1 kN).
A la figure 60 est représentée la courbe charge-déplacement en tête lors du premier cycle,

tandis que les répartitions de contraintes le long du fût sont indiquées figure 61. Les chemins
de contraintes, à diverses profondeurs (couche 1 en surface, et 5 près de la pointe), sont
visibles sur la figure 62. Les amplitudes cycliques sont importantes. Les caractères de ces
cycles sont indiqués au tableau 11.
Figure 59 . Répartition entre forces en tête et en pointe pour le pieu modélisé en compression
monotone (6.1 kN). Pointe/tête = 27.9 %.
347
Figure 60. Modélisation de l’essai de pieu centrifugé en compression cyclique CO4TO7.
Courbe charge-déplacement en tête lors du premier cycle (0, 1.2 kN, 3.6 kN, 1.2 kN, 2.4 kN).
La recharge de 1.2 à 2.4 kN est pratiquement confondue avec la décharge correspondante.

Distribution de contraintes a) normale b) de cisaillement le long du fût, au cours du premier
cycle de chargement.
Tableau 11. Modélisation de l’essai de pieu centrifugé en compression cyclique CO4TO7.

Caractères du premier cycle de chargement, à diverses profondeurs.
348
Chemins de contraintes, à diverses profondeurs (couche 1 en surface, et 5 près de la pointe).
Figure 63. Modélisation de l’essai de pieu modèle centrifugé en compression cyclique

CO4TO7 ; Force cyclique moyenne-déplacement cyclique moyen en tête de pieu résultant du
pseudo-fluage cyclique au bout de 2000 cycles.
On atteint ainsi un déplacement cyclique moyen de 0.50 mm au bout de 2000 cycles. Cette
valeur est à comparer à la valeur relative au pieu en traction cyclique, 1.54 mm. La figure 64
donne la distribution de contraintes le long du pieu au terme des 2000 cycles. La comparaison
avec le pieu en traction cyclique est intéressante : Le niveau de contrainte normale près de la
pointe décroit entre le 1er cycle et le dernier, mais reste important (≈ 150 kPa). D’où un
frottement latéral qui reste conséquent après cycles dans le cas d’une sollicitation en
compression. Cette analyse du pseudo-fluage cyclique relatif aux pieux est rustique, il s’agit à
ce stade d’une méthodologie.
349
Figure 64. Modélisation de l’essai de pieu modèle centrifugé en compression cyclique
CO4TO7. Distribution de contraintes a) normale b) de cisaillement le long du pieu au terme
des 2000 cycles.
Chapitre 5. Calcul de pieux réels chargés cycliquement, pour les projets de génie civil
Le calcul des pieux sous sollicitation cyclique mobilise les géotechniciens depuis de
nombreuses années, que ce soit pour le transfert de charge, pour la capacité limite, et pour la
relation charge déplacement en tête. Les méthodes des courbes t-z et p-y ont la faveur des
praticiens, en raison de leur simplicité. Le calcul pas à pas des cycles semble rédhibitoire au
vu du temps de calcul, mais aussi de la fiabilité des lois constitutives cycliques traitées pas à
pas, au-delà de quelques cycles. Le pseudo-fluage cyclique (le nombre de cycles agit comme
un temps fictif) s’impose pour traiter le grand nombre de cycles. Encore s’agit-il d’utiliser des
lois constitutives représentatives du phénomène. Les variables cycliques moyennes sont tout
naturellement les variables du pseudo-fluage cyclique.
Pour avoir quelques chances d’être représentatif, tout calcul doit impérativement modéliser
les mécanismes essentiels intervenant dans le sol et à l’interface sol-pieu. L’interface sol-pieu
est le centre névralgique du transfert des contraintes entre le pieu et le sol. De plus, s’agissant
de pieux sous chargement axial cyclique, la priorité doit être donnée à l’interface sol-pieu
sachant que la diffusion des sur-contraintes dues au chargement du pieu se diffuse en 1/r à
partir du fût, r étant la distance à l’axe du pieu. Par ailleurs, le sol contigu au pieu confine le
pieu (et l’interface), et ce confinement se manifeste par la rigidité normale k du massif de sol,
2G
de l’ordre de k = ( G module de cisaillement du sol, et R rayon du pieu). Si bien que les
R
chemins de cisaillement cycliques mobilisés le long du pieu sont à peu de chose près des
chemins à rigidité normale imposée (CNS). Cette rigidité normale est généralement faible,
mais présente, pour les pieux courants, de l’ordre de 1000 kPa/mm. Nous avons observé,
d’après nos essais cycliques CNS, que ces chemins étaient toujours contractants, conduisant à
une diminution de contrainte normale agissant sur le pieu. Ainsi, la dégradation du frottement,
souvent (mais heureusement de moins en moins) considérée comme diminution du coefficient
de frottement, correspond en fait à une réduction du niveau de contrainte normale.
La question de la modélisation du mode d’installation du pieu se pose bien évidemment. Elle
est importante, et nos 2 calculs de pieux modèles centrifugés montrent, par exemple, que ce
mode d’installation génère un frottement négatif influant sur la suite du chargement.
350
Après ces constatations, nous proposons la méthodologie suivante, impliquant un nombre
minimum d’essais, pour modéliser le comportement d’un pieu sous chargement axial
cyclique.
- Etat initial résultant du mode d’installation :

o Dans le cas du pieu moulé, la solution la plus simple consiste à remplace le sol
par le matériau pieu dans son emprise, le matériau pieu étant liquide en début
d’opération, coefficient de Poisson voisin de ½.
o Dans le cas du pieu foncé ou battu, on sait qu’il existe du frottement négatif
avant tout chargement. Nos calculs de pieux modèles centrifugés ont
opportunément mis en évidence un frottement négatif considérable. Pourquoi
ne pas modéliser le fonçage ou le battage par une excursion en accélération
verticale sur le modèle (1g → quelques g → 1g) ?
- Caractérisation du chargement du pieu, en tête et localement :
o Il s’agit d’analyser le programme de chargement en tête probable du pieu, en le
simplifiant sous forme de séries de cycles (valeur moyenne, amplitude), étant
entendu que les séries les plus sévères seront les seules modélisées.
o On privilégie délibérément les phénomènes d’interface lors de la modélisation
cyclique ultérieure, c'est-à-dire que l’on utilisera un outil éléments finis ou
différences finies mettant en œuvre une élasto-plasticité classique pour le sol.
o On réalise un calcul préliminaire du pieu, avec quelques cycles traités pas à
pas, afin de déterminer approximativement les paramètres cycliques locaux
(contrainte normale cyclique moyenne, cisaillement cyclique moyen,
amplitude cyclique).
- Essais de sol et d’interface sol-structure :
o On suppose avoir à disposition les essais courants de sol, ou les informations
équivalentes, censés fournir les distributions du poids volumique du sol, de la
teneur en eau, de la résistance du sol, de la rigidité du sol (Ep, ou G
notamment).
o L’information concernant les paramètres cycliques locaux (§ précédent)
permet de réduire le nombre, et de mieux cibler les paramètres, des essais de
cisaillement direct cyclique caractérisant le pieu, le sol et le chargement du
pieu. Il s’agira de quelques (3 à 5 ?) essais de cisaillement direct cycliques à
rigidité normale imposée (CNS), selon les paramètres cycliques précédemment
définis. La chute de contrainte normale au cours des cycles, et en fonction de la
profondeur, est une information de première importance à extraire de ces
essais. L’interpolation est un outil de grande qualité à manier sans modération,
pour compléter ces données.
- Modélisation proprement dite du comportement du pieu sous chargement axial
cyclique, à l’aide d’un logiciel structural éléments finis ou différences finies :
o 0. Phase d’initialisation des contraintes géostatiques
o 1. Phase de modélisation du mode d’installation du pieu
o 2. Pour chaque série de cycles, Parcours pas à pas d’1 ou 2 cycles de
chargement en tête
o 3. Extraction des paramètres cycliques locaux
o 4. Formulation des courbes de chute de contrainte normale fonction du nombre
N de cycles de la série, relatives à ces conditions locales (Figure 65)
o 5. Prescription de ces chutes de contrainte normale locale dans le logiciel
structural, pour modéliser les N cycles de cette série.
o 6. Si cette série de cycles est la dernière, passer à 7, sinon retourner à 2.
351
Figure 65. Résultats typiques d’essais d’interface sol-structure sur chemin CNS directement
utilisable pour la modélisation de la chute de contrainte normale le long d’un pieu chargé
cycliquement. ID0 = 90 %, ηcm0 = 0.35, k = 1000 kPa/mm.
Conclusions générales et perspectives
Nous sommes partis d’une petite somme de connaissances internationales concernant le

comportement des interfaces sol-structure sous sollicitations cycliques. La résistance
d’interface n’est pas la seule information utile : les possibilités de dilatance et/ou contractance
d’interface sont tout aussi utiles. Nous avons donc développé une modification de machine de
cisaillement direct apte à réaliser des essais de cisaillement direct à rigidité normale imposée,
pour des sollicitations cycliques à grand nombre de cycles (le plus souvent 10000).
Cette machine nous a permis de mener à bien plusieurs séries d’essais. Une série cohérente
sable de Fontainebleau-plaque rugueuse à contrainte normale constante et à rigidité normale
imposée, une série cohérente aussi, mais avec un nombre réduit d’essais, pour plaque
rugueuse, et enfin une série encore plus limitée relative au sable de Loon-Plage, site sur lequel
des essais de pieu grandeur réelle sous sollicitation cyclique sont disponibles dans le cadre
SOLCYP (ANR et Programme National). Après tous ces essais, une analyse granulométrique
de la zone d’interface a été faite, mais non exploitée. Les essais plaque lisse et Loon-Plage
n’ont pas été exploités non plus. L’échantillon testé a été considéré comme formé de 2
couches, l’interface, de faible épaisseur d’une part, le tampon pseudo-oedométrique d’autre
part, n’intervenant que lors des variations de contraintes normales. Ces essais ont été
interprétés dans un premier temps en termes de variables cycliques moyennes, et de poids
volumique d’interface sous contrainte.
En référence au modèle élasto-visco-plastique de Perzyna, nous avons modélisé
analytiquement, d’une manière phénoménologique, les essais sable-plaque rugueuse sur
chemin CNL (dits identification) en traitant d’abord les densités faible (ID0 = 30 %) et forte
(ID0 = 90 %), les cisaillements cycliques moyens nul, moyen et élevé, et plusieurs niveau
d’amplitude cyclique. Le paramètre de mémoire est le poids volumique d’interface sous
contrainte (normale) ou le déplacement relatif normal d’interface. Les autres chemins CNL
ont été générés par interpolation (rhéologique). Nous avons également modélisé les essais sur
chemins différents (dits validation), en intégrant par pas analytiques pré-intégrés. Il s’est agit
de chemins à amplitudes successives variées, et de chemins CNS, ceux-ci faisant appel aussi
au comportement pseudo-oedométrique. L’effet de la taille du pas d’intégration a été
examiné. Les comparaisons expérience-modèle sont généralement assez satisfaisantes.
352
Au terme de ces essais et de leur modélisation constitutive, nous nous sommes consacrés à la
modélisation structurale en relation avec notre sujet, en utilisant le logiciel éléments finis
PLAXIS 2D et sa loi constitutive opar défaut : l’élasto-plasticité à 5 paramètres.
Nous avons d’abord modélisé l’essai de cisaillement lui-même, en 2D (malheureusement pas
en 3D), afin de détecter, lors d’une sollicitation monotone, les hétérogénéités de cisaillement,
et de valider notre hypothèse de 2 couches globalement homogènes dans la boîte. Nous avons
également modélisé les chemins CNL et CNS. Ces comparaisons sont assez satisfaisante.
Puis nous avons abordé la modélisation de pieux sous chargement axial cyclique. Il s’agit
d’essais de pieux modèles centrifugés menés à bien par l’IFSTTAR à Nantes. Nous avons
traité un cas de traction cyclique et un cas de compression cyclique, d’une manière que nous
qualifions de rustique, disons « avec les moyens du bord ». Il est clair que le logiciel PLAXIS,
pas plus que d’autres, ne possède pas la faculté de modéliser le grand nombre de cycles, avec
les lois constitutives disponibles en bibliothèque. D’autre part, nous n’avons pas eu le temps
matériel d’implémenter notre loi d’interface en tant que loi utilisateur. Nous avons donc du
prendre une voie détournée. Nous avons travaillé en dialogue entre EXCEL et PLAXIS. Nous
avons modélisé la chute de contrainte normale locale sur chemin CNS, issue de nos résultats
sur le nombre de cycles conduisant à la rupture sur chemin CNS. Nous avons opéré par cluster
le long du pieu, en imposant une variation de volume locale dans les dits clusters, d’où un
déplacement irréversible (pseudo-fluage cyclique) en tête de pieu, à partir du déplacement
moyen en tête au cours des premiers cycles.
De nombreuses perspectives de travaux futurs sont ouvertes par nos travaux. Au plan
expérimental, nous n’avons traité que des cycles d’amplitude relativement faible. Il y aurait
lieu de compléter nos essais par des cycles plus amples. Ces essais seraient de très longue
durée, sachant que nos essais à 10000 cycles duraient déjà environ une semaine. Par ailleurs
nous n’avons pas exploité nos essais avec plaque lisse, ni les résultats de l’évolution de
granulométrie du sable de Fontainebleau due au cisaillement cyclique. Il est vrai que chaque
échantillon de sable de la zone d’interface représente une très faible quantité (quelques
grammes). En termes de modélisation constitutive, il y aurait sans doute lieu d’opérer
quelques simplifications (réduction du nombre de paramètres) en reliant des coefficients
vraisemblablement corrélés. Un travail très important pourrait être effectué sur le plan de la
modélisation structurale des pieux sollicités cycliquement. Maillage plus élaboré, travailler au
niveau des points d’intégration plutôt qu’au niveau de quelques clusters, automatiser les
échanges entre PLAXIS et les données du pseudo-fluage cyclique en reprogrammant la loi
d’interface cyclique en langage C, et enfin utiliser des chemins CNS réellement intégrés au
lieu de la simplification apportée par Ncrit, et un seul pas de pseudo-fluage cyclique.
L’automatisation des échanges est possible, car elle a été réalisée dans la thèse de S.
Levasseur
Annexes
L’annexe A contient le détail de calculs analytiques relatifs au modèle de cisaillement direct

monotone.
L’annexe B contient les résultats expérimentaux de tous les essais de cisaillement direct
monotones et cycliques réalisés, ainsi que leur modélisation constitutive.
353

Behaviour of Soil-Structure Interfaces Subjected To A Large Number of Cycles, Application To Pile

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Behaviour of Soil-Structure Interfaces Subjected To A Large Number of Cycles, Application To Pile

Uploaded by

Copyright:

Available Formats

THÈSE

Pour obtenir le grade de

Thèse dirigée par Pierre FORAY et

préparée au sein du Laboratoire 3S-R

Thèse soutenue publiquement le 28.02.2013,

A series of monotonic and cyclic 2D direct shear tests on sand (Fontainebleau)-rough/smooth

• CNL Constant Normal Stress condition

1 Bibliographic study of soil-structure interfaces subjected to cyclic loading 3

3 Constitutive modeling of direct shear tests 157

4 Finite element modeling 214

5 Calculation of real piles subjected to cyclic loading, for civil engineering

General conclusions and perspectives 248

A Analytical model of monotonic interface tests 257

B Results for Monotonic CNL and CNS tests 262

C Cyclic interface models 283

D Cyclic CNL tests 288

F Résumé français étendu de la Thèse 301

Soil-structure interfaces are commonly encountered in many geotechnical engineering prob-

1.2 Soil-structure interface

Soil-structure interface is the result of a phenomenon induced by the discontinuity and

τs = Kσn′ tanδ (1.2)

So far, the development of accurate understanding of the mechanical response of interfaces

1.3.1 Interface shear devices

Smooth plate L = D50

sheear stress at contact-face /kPa

1.3.3 Interface thickness

1.4 Cyclic CNL interface shear tests

1.5 CNS interface shear tests

1.5.1 Several shear paths

Nevertheless, from pull-out test of a model-pile embedded in sand performed by Puech et

1.5.2 Definition of Constant Normal Stiffness (CNS)

A laboratory test of pile-soil interface can simply be interpreted as an interface under

structure structure structure

1.5.3 Cyclic CNS interface shear tests

shear stress, W (kPa)

rough surface smooth surface

shear displacement, [w] (mm) shear displacement, [w] (mm)

rough surface smooth surface

shear displacement, [w] (mm) shear displacement, [w] (mm)

rough surface smooth surface

normal stress, Vn (kPa) normal stress, Vn (kPa)

p = (σ1 + 2σ3 )/3 (1.9)

εq = 2/3(ε1 − 2ε3 ) (1.12)

where εacc acc

1.6.1 Influence of the strain amplitude

1.6.2 Influence of initial density, ID0

1.6.3 Influence of of loading frequency

1.6.4 Influence of of the average stress

A = active interface area

1.7.2 Interface constitutive models

[u] = {[w] [u]}T (1.19)

The time derivatives of these state variables are then expressed:

ṫ = {τ̇ σ˙n }T (1.20)

ṫ = d (loading state) [u̇] (1.22)

The matrix d containing the stiffness of interface is expressed by:

• Interface shear test at constant σn , (CNL);

• Interface shear test at constant volume (CV);

• Interface shear test at constant normal stiffness ke , e ⇔ external, (CNS);

(a) (b) (c)

1 ks2 (1 − α) αks kn tan δ

1.7.3 Incremental models (Implicit in the sense of Wichtmann(2005))

and the corresponding incremental response:

where Wi are interpolation functions (e.g., polynomial, trigonometric).